STRATEGIES FOR IMPROVING THE

DETECTION ACCURACY OF

COMPUTERIZED MACHINE VISION

CONSIDERING SPATIAL APPLICATIONS

Mincheng Piao*

• School of Computer Science of South-Central Minzu University, Wuhan, Hubei,

430074, China

•E-mail: piaomincheng@163.com

Meng Song

• School of Textile Science and Engineering of Wuhan Textile University, Wuhan,

Hubei, 430200, China

Reception: 27 December 2023 | Acceptance: 24 January 2024 | Publication: 13 February 2024

Suggested citation:

Piao, M. and Song, M. (2024). Strategies for improving the detection

accuracy of computerized machine vision considering spatial

applications. 3C TIC. Cuadernos de desarrollo aplicados a las TIC, 13(1), 76-94.

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

ABSTRACT

In this paper, strategies in image preprocessing, hardware composition and detection

methods are considered to improve computerized machine vision detection accuracy.

First, image preprocessing and image enhancement are performed to improve the

quality of the input image. Second, the hardware composition of the computer vision

online inspection system is optimized by focusing on the light source selection and the

performance of the image acquisition card in spatial applications. Combined with

spatial application calculations, methods such as frequency domain method and

Canny operator are used in order to improve the accuracy of machine vision

detection. Finally, in the same test environment, the machine vision detection requires

only 400MB and the detection accuracy ranges from 85.13% to 99.42%. With these

comprehensive strategies, this paper provides a comprehensive and effective

approach for computerized machine vision detection in spatial applications to improve

detection accuracy and meet demanding application scenarios.

KEYWORDS

Image preprocessing; machine vision detection; hardware composition; frequency

domain method; Canny operator

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

INDEX

ABSTRACT .....................................................................................................................2

KEYWORDS ...................................................................................................................2

1. INTRODUCTION .......................................................................................................4

2. LITERATURE REVIEW .............................................................................................4

3. COMPUTERIZED IMAGE PREPROCESSING ........................................................6

3.1. Pretreatment process .........................................................................................6

3.2. Image Enhancement ..........................................................................................7

4. HARDWARE COMPOSITION OF COMPUTERIZED MACHINE VISION ONLINE

INSPECTION SYSTEM ............................................................................................8

4.1. Composition of computer vision online workpiece inspection system ................8

4.2. Detection light source selection considering spatial applications .......................8

4.3. Image Acquisition Card ......................................................................................9

5. CALCULATION PROCESS OF MACHINE VISION INSPECTION METHOD .......10

5.1. Image Segmentation ........................................................................................10

5.2. Spatial filtering calculations ..............................................................................11

5.2.1. Frequency domain method ........................................................................11

5.2.2. anny operators ..........................................................................................12

5.3. Detection process ............................................................................................12

6. DETECTION ACCURACY IMPROVEMENT STRATEGY ANALYSIS ...................13

6.1. Detection errors ................................................................................................13

6.2. Comparative validation .....................................................................................14

6.3. Algorithm performance validation .....................................................................15

6.4. Comparison of the amount of detection errors .................................................16

7. CONCLUSION ........................................................................................................17

ABOUT THE AUTHORS ...............................................................................................18

REFERENCES ..............................................................................................................18

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

1. INTRODUCTION

With the continuous progress of science and technology, computerized testing has

developed rapidly, and the categories of computerized products are getting richer and

richer, and the quality requirements for testing are getting higher and higher [1]. In

machine management, the accuracy of the detection results is very critical [2]. Due to

the influence of factors such as the failure of traditional detection equipment or

operational errors in the detection process, the detection results have some errors [3].

The spatial environment has a serious impact on the results of computer testing,

which not only causes economic benefit loss to the relevant enterprises, but also

seriously affects the competition of the computer industry in the international arena [4].

At present, most of the computer testing relies on manual to detect and identify the

fault point, however, the manual detection method has limited detection accuracy,

detection speed is not high, the test results are easy to be affected by the subjective

factors of manual, high labor costs [5]. Thus making the computer in operation there

are potential quality hazards, which directly leads to the inefficiency of machine vision

inspection, reducing the competitiveness of the computer industry, and ultimately may

face the end of being eliminated from the market [6]. Therefore, how to carry out rapid

and effective detection of computers to improve the detection accuracy is an urgent

problem to be solved.

Although the existing computerized detection methods have achieved some

success in various aspects, the influence of spatial environmental factors has not

been considered. In this paper, the input image is optimized and its quality is

enhanced through computer image preprocessing. Various components required to

construct an online workpiece inspection system for computer vision, especially in

spatial applications, light sources adapted to the spatial environment are considered

to improve the clarity and contrast of the workpiece. In the frequency-domain method,

the fluctuation characteristics specific to spatial applications are emphasized to better

adapt to the spatial environment. The application of Canny operator further improves

the sensitivity to edges and enhances the detection effect. The accuracy

enhancement strategy includes optimizing the preprocessing process, choosing a

hardware composition adapted to the spatial environment, adjusting the light source,

and improving the spatial application calculation method.

2. LITERATURE REVIEW

Sangirardi, M et al. showed that detecting the onset of structural damage and its

gradual evolution is essential for the assessment and maintenance of the built

environment, and in their study presented the application of a computer vision based

structural health monitoring methodology for shaking table investigations [7]. Liu, G et

al. addressed the fact that traffic management systems can capture large amounts of

video data and use video processing techniques to detect and monitor traffic

accidents. The collected data is traditionally forwarded to a traffic management center

for in-depth analysis, which can exacerbate the problem of network paths to the traffic

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

management center. In the study, it is pointed out to utilize edge computing to equip

edge nodes close to the camera with computational resources [8]. Wu, Z et al.

proposed a computer vision based weed detection method, which explores the

solution to the weed detection problem in terms of both traditional image processing

and deep learning. Various weed detection methods in recent years were

summarized, the advantages and disadvantages were analyzed, and relevant plant

leaves, weed datasets, and weeding machines were introduced [9]. Li, Z et al.

proposed a static software defect detection system design based on big data

technology, which aimed to optimize the design of the traditional system. By predicting

potentially defective program modules, the system design improves the hardware and

software structure and achieves efficient allocation of testing resources, thus

improving the quality of software products [10]. Lee, H et al. proposed a feature

descriptor-based computer vision method for detecting rail corrugations to achieve

automatic differentiation between corrugated and normal surfaces. The authors

extracted seven features and combined them into a feature vector for constructing a

support vector machine. The results show that the method is more effective than

References in the recognition of corrugated images [11].

Qiao, W et al. redesigned the structure of DenseNet and improved it by adding the

Expected Maximum Attention (EMA) module after the last pooling layer. The EMA

module plays a significant role in bridge damage feature extraction. In addition, a loss

function considering pixel connectivity is used in the paper, which shows good results

in reducing the breakpoints of fracture prediction as well as improving the accuracy of

fracture prediction, and the application in computer vision inspection helps to improve

the accuracy and precision of bridge damage [12]. Shi, H et al. worked on the quality

of steel wire ropes in lifting equipment, and used an industrial video camera to acquire

infrared images of steel wire ropes. The wire rope contour was extracted by Canny

edge detection and the diameter of the wire rope was corrected by one-dimensional

measurements and directional fitting. The combination of computer vision inspection

is expected to play an important role in improving the accuracy and efficiency of wire

rope quality inspection [13]. Ljubovic, V et al. introduced a novel approach to source

code repository construction by storing each editing event in the program source as a

new commit, resulting in an ultra-fine-grained source code repository. Machine

learning techniques were applied to detect suspicious behavior, thus significantly

improving the performance of traditional plagiarism detection tools [14]. Roy, S. D et

al. used a variety of feature extractors, including texture features such as SIFT, SURF,

and ORB, as well as statistical features such as Haralick texture features, to form a

dataset containing 782 features. Then, by stacking these features using multiple

machine learning classifiers and using Pearson's correlation coefficient for feature

selection, a dataset containing four features was finally generated for classification.

This study combined the advantages of computer vision and machine learning to

achieve good performance in feature extraction and classification tasks [15]. Huang, H

et al. utilized a combination of computer vision, machine learning, and edge

computing to provide an efficient and accurate solution for the citrus detection task. To

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

facilitate the deployment of the model, a pruning approach was used to reduce the

computational effort and parameters of the model [16].

3. COMPUTERIZED IMAGE PREPROCESSING

3.1. PRETREATMENT PROCESS

Considering the conditions of spatial factors, when computer detection is carried

out with the help of computer vision technology, it is necessary to pay attention to the

image pre-processing technology, which is closely related to the subsequent image

processing and analysis [17]. Figure 1 shows the image preprocessing process, to

first extract the relevant information data of the image, and then effectively integrate

the image preprocessing technology and template technology, thus reducing the

technical difficulty of the actual monitoring. Based on the actual technical

requirements, during the implementation of image pre-processing, the efficiency of the

use of images should be improved. After completing the pre-visualization process,

carry out the two-dimensional numerical execution of the marginalization extraction

operation, effectively input the whole frame of image data, extract the image edges,

and clarify the key nodes of the processing technology, so as to make it meet the

stability requirements. In addition, in the course of practice, the previsualization

processing should be performed several times.

Figure 1. Image preprocessing process

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

3.2. IMAGE ENHANCEMENT

Image enhancement techniques are spatial domain processing method and

frequency domain processing method, the spatial domain processing method mainly

focuses on direct operations on image pixels in the spatial domain [18]. In this paper

image enhancement technique using spatial domain method is described by the

following equation:

(1)

Where is the image before processing,

denotes the image after

processing and

is the spatial operation function. The frequency domain

processing method of image enhancement is to process the transformed values of an

image in some transform domain, usually the frequency domain, by performing some

kind of operation on the transformed values of the image and then transforming back

to the spatial domain. It is an indirect processing method with the following process:

1. First input the original image for positive transformation.

2. After the positive transformation,

is obtained, which is corrected to

obtain .

3. After transforming

, inverse transform is performed and the enhanced

image is output.

The above mathematical description is as follows:

(2)

(3)

(4)

Where denotes some frequency domain positive transform,

denotes

the inverse transform of that frequency domain transform

is the result of the

frequency domain positive transform of the original image, is the

number of positives in the frequency domain

is the result of the correction,

and is the result of the

inverse transform, which is the enhanced

image.

g(x,y)=f(x,y)⋅h(x,y)

f(x,y)

g(x,y)

h(x,y)

f(x,y)

F(μ,v)

H(μ,v)

G(μ,v)

g(x,y)

F(μ,v) = T{f(x,y)}

G(μ,v)=H(μ,v)⋅F(μ,v)

g(x,y)=T−1{G(μ,v)}

T{}

T−1{}

F(μ,v)

g(x,y)

H(μ,v)

G(μ,v)

g(x,y)

G(μ,v)

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

4. HARDWARE COMPOSITION OF COMPUTERIZED

MACHINE VISION ONLINE INSPECTION SYSTEM

4.1. COMPOSITION OF COMPUTER VISION ONLINE

WORKPIECE INSPECTION SYSTEM

Computer vision online workpiece inspection system belongs to the application of

computer vision system in detecting parts, and the system is required to be able to

accurately observe the target and make effective decisions on which parts can pass

the inspection and which need to be discarded [19-20]. The computer vision online

workpiece inspection system is shown in Fig. 2, which should include several parts

such as light source, optical system, CCD camera, image acquisition card, image

processing module and fast and accurate actuator.

Figure 2. Structure of computer vision on-line workpiece inspection system

4.2. DETECTION LIGHT SOURCE SELECTION CONSIDERING

SPATIAL APPLICATIONS

In computer vision application systems, a good light source and illumination

scheme is often the key to the success or failure of the whole system, and plays a

very important role [21]. The cooperation of light source and illumination scheme

should highlight the amount of object features as much as possible, in the part of the

object that needs to be detected, and those unimportant parts should be as much as

possible to produce a clear difference between the parts to increase the contrast. It

should also ensure that the overall brightness is sufficient, and changes in the position

of the object should not affect the quality of the image. Transmitted light and reflected

light are generally used in vision inspection applications. For the reflected light

situation should fully consider the relative position of the light source and optical lens,

the texture of the object surface, the geometry of the object and other elements. The

choice of light source equipment must be consistent with the desired geometry,

illumination brightness, uniformity, the spectral characteristics of the light emitted must

also meet the actual requirements, but also consider the luminous efficiency and

service life of the light source. In short, different forms of light sources should be

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

selected and designed according to the actual task in order to achieve the best state

of object imaging.

Table 1 shows the classification of commonly used light sources. LED light source

has the advantages of free shape, long service life, fast response speed, as well as

free choice of color and low comprehensive running cost, which makes it the most

suitable dedicated light source for computer vision inspection system applications.

Table 1. Classification of common light sources

Space due to some of the measured object surface reflection phenomenon is very

serious, you can use diffuse reflection light illumination method for illumination. Both to

avoid the appearance of reflections, but also to facilitate the subsequent various

image processing algorithms. Lambert’s law states that the effect of diffuse reflection

is related to the orientation of the surface with respect to the light source, that is:

(5)

Where

is the brightness of a point on a visible surface caused by diffuse

reflection. is the brightness caused by incident light from a point source.

is the

diffuse reflection coefficient, which takes values between 0 and 1 and varies with the

material of the object. Is the angle of incidence between the visible surface in the

direction normal to and the point light source in the direction

, which should be

between 0° and 90°.

4.3. IMAGE ACQUISITION CARD

At present, there are many types of image acquisition cards, according to different

classification methods, there are black and white image and color image acquisition

cards, analog and digital signal acquisition cards, composite signals and RGB

component signal input acquisition cards. Figure 3 for the image acquisition card

structure framework, image acquisition card generally has the following functional

modules:

Image signal reception and A/D conversion module, responsible for image

signal amplification and digitization.

Complex

design

Service

life

Temperature

effect

Degree of

stability Costs Luminance

Fuorescent

tube Lower General General Differ from Lower Lower

Halogen

lamp + fiber

optic conduit

General Differ

from Differ from General Above

average

Above

average

LED light

source

Above

average Good Lower Good General General

Id=Ip⋅Kd⋅cos θ

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

2. Camera control input and output interfaces, mainly responsible for coordinating

the camera for synchronization or to achieve asynchronous reset photo, timed

photo and so on.

Bus interface, responsible for high-speed output of digital data through the

computer's internal bus, generally PCI interface, the transmission rate can be

as high as 130Mbps, fully capable of high-precision image transmission in real

time, and occupies less CPU time.

4. Display module, responsible for high-quality image display in real time.

5. Communication interface, responsible for communication.

When selecting the image acquisition card, the main consideration should be the

functional requirements of the system, the image acquisition accuracy and the

matching of the output signal with the camera and other factors.

Figure 3. Acquisition card structure framework

5. CALCULATION PROCESS OF MACHINE VISION

INSPECTION METHOD

5.1. IMAGE SEGMENTATION

In computers, including the four primary colors cyan, magenta, yellow and black,

which are used in the usual space, the accuracy detection algorithm is based on black

in order to calculate the distance between each monochrome color to the black

baseline [22]. Therefore, it is necessary to convert the acquired RGB mode image to

CMYK model, and the conversion is shown in Eq:

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

(6)

In the RGB to CMYK conversion formula, the black color separation (K) is

calculated from red (R), green (G) and blue (B). Specifically, the value of the black

color separation is:

(7)

5.2. SPATIAL FILTERING CALCULATIONS

Spatial filters can be subdivided into linear and nonlinear filters depending on

whether the processing is linear in the image system [23]. A linear spatial filter is at

any point in the image, and the sum of the product of the coefficients of this type

of filter and the corresponding pixel values of the region scanned by the mask is the

filter response at that point.

Nonlinear spatial filtering is the same as linear spatial filtering in that it scans the

image to be processed with a pre-defined mask, but it cannot be used directly with the

filter coefficients, and the product and sum of the corresponding pixel values of the

area city scanned by the mask to obtain the response value at this point. Nonlinear

spatial filters include median filters, statistical ordering filters, and so on. The median

filter can protect the edge pixel points of the image, and has a low degree of

smoothing and blurring compared with the linear spatial filter of the same size. The

nonlinear spatial filter filtering mechanism is to replace the pixel value at the origin

with the maximum value pixel, or the minimum value pixel after sorting the pixel points

in the neighborhood.

5.2.1. FREQUENCY DOMAIN METHOD

An ideal low-pass filter will pass all frequencies without attenuation in a circle

centered at the origin and radiused at . The functional expression is given in

equation (8):

(8)

where is the cutoff frequency, i.e:

(9)

Where is the size of the image to fill the image with complementary zeros.

−

K= min(C,M,Y)

(x,y)

g(x,y)

(u,v) =

{1D(u,v)≤D0

0D(u,v)>D0

D(u,v)=[(u−P/2)2+ (v−Q/2)2]1/2

P,Q

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

5.2.2. ANNY OPERATORS

Canny operator algorithm has low error rate, no pseudo response and can detect

real edges when dealing with edge detection of images. Firstly, the smoothing filter of

the image to be processed is carried out, here the smoothing filter is a Gaussian filter

and the gradient is calculated, and the intensity and direction of the edge pixel points

are estimated using the obtained gradient size and direction. Then the non-maximum

value suppression method is used for edge refinement, and the edge direction of the

center point of the neighborhood is quantized into horizontal, vertical, and two

diagonal directions, and the image edge direction is determined by the edge normal

direction. Finally, to reduce the detection of pseudo edge points, the sub-detection of

the true edge of the image, using the lagged intrusion value method, that is, in the

range of Min value to take out the high value and low Que value, and set the ratio of

2:1 or 3:1.

The image to be processed is and the Gaussian function is , then it

can be expressed as:

(10)

Where is the standard deviation.

After performing Gaussian filtering the image is which can be expressed as:

(11)

5.3. DETECTION PROCESS

Based on the above, the software detection process is shown in Fig. 2, where the

user can specify the model to be used in image processing by selecting the model,

involving different algorithms or techniques. Subsequently, in the image enhancement

stage, the visual quality of the image can be improved to make it more suitable for

analysis for spatial applications. Next, in the image feature extraction stage,

meaningful features are involved to be extracted from the image that are essential for

improving the accuracy of computerized machine vision detection for spatial

applications. In the import modeling computation step, a machine learning model is

introduced for image processing. In addition, the user can ensure effective

communication between the image processing device and the computer by

connecting the computer. For the wide only image set to be detected, it may refer to a

wide set of images prepared to be detected. Subsequently, image acquisition and

image acquisition involves acquiring image data by means of a camera, a scanner,

and the like. In a filtering and noise reduction step, noise is removed from the image

by using filtering techniques to improve the image quality. Next, image segmentation

and edge extraction involves segmenting the image into different regions and

detecting edges of objects in the image. The image preprocessing stage may then

fs(x,y)

G(x,y)

G(x,y)=e

−

x2+y2

2δ2

fs(x,y)

fs(x,y)=G(x,y)*f(x,y)

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

include further enhancement and preparation of the image to ensure adaptation to

subsequent analysis needs. Upon completion of the image processing, the user may

be given the option to display the data to view the processed image or data and store

it for future reference by saving the data. Finally, an exit option is used to end the

image processing program or system. The entire process also includes a

computerized accuracy check, which is used to assess the accuracy of the computer

algorithms during processing.

Figure 4. Software testing process

6. DETECTION ACCURACY IMPROVEMENT STRATEGY

ANALYSIS

6.1. DETECTION ERRORS

In order to evaluate the performance of the proposed machine vision inspection

enhancement strategy on different inspection samples, the variation of the detection

error is first analyzed and Table 2 shows the variation of the detection error. The

detection error varies between samples, but generally stays at a low level. The error

ranges from 1.4% to 3.4%, which indicates that the proposed strategy can achieve

accurate detection in most cases. There are some differences between the actual

data and the test data, and the errors are mainly due to the deviation between these

two. However, this deviation is relatively small in most cases. In summary, the

proposed strategy for improving the accuracy of machine vision detection shows high

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

accuracy under multiple detection samples, and the error is kept within a reasonable

range, which helps to improve the performance of computerized machine vision in

spatial applications.

Table 2. Detection errors

6.2. COMPARATIVE VALIDATION

In order to examine the accuracy of computerized machine vision data detection in

spatial application scenarios, four different methods are compared, namely the

proposed strategy, deep learning, artificial intelligence, and virtual reality. The

performance of these methods in dealing with complex spatial data is evaluated and

compared by analyzing 20 different detections. Figure 5 shows a comparison of the

detection accuracy of the different methods, and the accuracy of the proposed

computerized machine data detection for considering spatial applications ranges from

85.13% to 99.42%, which shows an extremely high detection accuracy. In particular,

99.42 is achieved in the number of detections 8, and on 97.54% for the number of

detections at 7. The proposed strategy demonstrates significant high accuracy. The

accuracy of deep learning ranges from 62.84% to 89.01%. Deep learning performed

poorly on most data points compared to the proposed strategy, but reached 89.01% at

15 times. As well as performing better on 16 times at 88.99%. The accuracy of

Artificial Intelligence ranges from 60.26% to 89.17%. Similar to Deep Learning,

Artificial Intelligence lagged behind the proposed strategy on most data points, but

had a high performance on 89.17% at the 3rd time and 87.24% at the 12th time. The

accuracy of Virtual Reality ranges from 50.97% to 79.84%, which is the lowest of the

four methods. This indicates that virtual reality faces some challenges in terms of

detection accuracy in cases where spatial applications are considered.

Test sample Actual data Test data Error

10 102.5 100.2 2.3

20 98.7 97.1 1.6

30 115.2 118.6 3.4

40 92.8 94.3 1.5

50 105.6 103.8 1.8

60 99.3 97.9 1.4

70 110.7 112.2 1.5

80 97.6 99.0 1.4

90 103.8 102.2 1.6

100 112.1 110.5 1.6

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

Figure 5. Comparison of detection accuracy

6.3. ALGORITHM PERFORMANCE VALIDATION

In this paper, a series of identical run time points, from 1 to 10 seconds, were

chosen to ensure a fair comparison. The computerized machine vision detection

algorithm and the traditional AI method were run separately and the CPU utilization at

each time point was recorded. Figure 6 shows the results of the algorithm

performance test, where the traditional AI method typically has higher resource

consumption than the machine vision detection algorithm for the same run time. For

example, the traditional AI method consumes 500 MB of memory in 10 seconds, while

the machine vision detection only requires 400 MB. Secondly, the traditional AI

method also typically exhibits higher CPU utilization. The resource consumption of

both methods tends to increase linearly as the runtime increases, but the traditional AI

method increases at a faster rate. Within 10 seconds, the CPU utilization of the

traditional AI method reaches 50%, while the machine vision detection algorithm uses

only 40%. In addition to this, the CPU utilization of both methods also shows a linear

growth trend, but the CPU utilization of machine vision detection grows at a slower

rate. For computerized machine vision inspection accuracy improvement strategies for

spatial applications, there is a need to weigh performance requirements and resource

availability. If computing resources are limited, algorithms or hardware need to be

optimized to reduce the growth rate of CPU usage to ensure system stability and

performance.

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

100

Figure 6. Algorithm performance verification

6.4. COMPARISON OF THE AMOUNT OF DETECTION ERRORS

In spatial applications, the performance of computerized machine vision detection

systems is critical and is directly related to the successful execution of the task and

the safety of the system. By comparing the detection quantities of machine learning,

artificial intelligence and virtual reality on different error types, key performance

bottlenecks can be identified and effective strategies can be developed for improving

the detection accuracy of the system. Table 3 shows the comparison of the amount of

errors detected by different methods, with machine learning detecting 25, AI 16, and

virtual reality 22. Robotic arm damage, as a high-frequency error type, requires

focusing on optimizing the algorithm, especially in machine learning for performance

improvement. For this problem, the introduction of more sophisticated feature

extraction and fusion methods can be considered to enhance the accurate detection

of the state of the robotic arm. Inclement weather or poor lighting conditions are a

prominent challenge, corresponding to a number of detections of 23, 15, and 20,

respectively. In this regard, adaptation to inclement weather and lighting conditions

should be optimized to ensure the reliability of the system in complex environments.

By taking these error types into account, a comprehensive enhancement strategy

including improving algorithm robustness, enhancing adaptation to specific

environments, and optimizing the quality of simulation of computer environments is

developed to improve the accuracy of computerized machine vision detection for

spatial applications.

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

101

Table 3. Comparison of error amount detected by different methods

7. CONCLUSION

The computerized machine vision inspection accuracy enhancement strategy

considering spatial applications proposed in this paper provides a reliable inspection

solution for spatial applications, and the conclusions are as follows:

The machine vision detection accuracy proposed in this paper for spatial

applications ranges from 85.13% to 99.42% while maintaining a low error

range of 1.4% to 3.4%. The proposed strategy performs better compared to

deep learning and traditional AI.

2. In terms of resource consumption, the traditional AI method consumes 500MB

of memory in 0 seconds, while machine vision detection only requires 400MB,

making the machine vision detection algorithm more efficient.

By comparing the amount of detection errors, the number of machine learning

detection is 25, artificial intelligence is 16, and virtual reality is 22. It reflects the

reliability and efficiency of the proposed method, which provides a strong

support for the development of machine vision detection systems in spatial

applications.

In conclusion, it is proved that the proposed method is not only better than the

traditional method in terms of accuracy, but also more efficient in terms of resource

Error type

Detection quantity comparison

Machine learning Artificial intelligence Virtual reality

19 8 12

Packet loss or

error in data

transmission

22 12 18

Low battery 8 4 6

Mechanical arm

failure 25 16 22

Image distortion or

artifact 8 7 6

Positioning

deviation 19 10 14

Algorithm

execution error 11 6 9

Bad weather or

poor light

conditions

23 15 20

Dangerous object

intrusion 6 2 4

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

102

consumption, which provides a strong support for the development of machine vision

detection systems in future space applications.

ABOUT THE AUTHORS

Mincheng Piao was born in Qingdao, Shandong, P.R. China, in 2002. He obtained

a bachelor's degree from South-Central Minzu University in China. I am currently

studying at the School of Computer Science, South-Central Minzu University. My main

research direction is Information fusion technology and Computer Vision.

Meng Song was born in Heze, Shandong, P.R. China, in 2002. she obtained a

bachelor's degree from Wuhan Textile University in China. I am currently studying at

the School of Textile Science and Engineering, Wuhan Textile University. My main

research direction is Textile Engineering and Material Analysis.

REFERENCES

(1) Sun, X., Liu, X., Yang, X., & Song, B. (2021). Computer-aided three-dimensional ceramic

product design. Computer-Aided Design and Applications, 19(S3), 97-107.

(2) Huang, J., Guixiong, L., & He, B. (2021). Fast semantic segmentation method for

machine vision inspection based on a fewer-parameters atrous convolution neural

network. PloS one, 16(2), e0246093.

(3) Althubaiti, A., Elasha, F., & Teixeira, J. A. (2022). Fault diagnosis and health

management of bearings in rotating equipment based on vibration analysis–a review.

Journal of Vibroengineering, 24(1), 46-74.

(4) Yin, J., Huang, S., Lei, L., & Yao, J. (2021). Intelligent monitoring method of short-

distance swimming physical function fatigue limit mobile calculation. Wireless

Communications and Mobile Computing, 2021, 1-6.

(5) González-Gaspar, P., Macías-Carballo, M., Cadena-Mejía, T., Landa-Jiménez, M. A.,

Montes-González, F. M., López-Meraz, M. L., ... & Morgado-Valle, C. (2021). Analixity:

An open source, low-cost analysis system for the elevated plus maze test, based on

computer vision techniques. Behavioural Processes, 193, 104539.

(6) Talaat, M., Arafa, I., & Metwally, H. M. B. (2020). Advanced automation system for

charging electric vehicles based on machine vision and finite element method. IET

Electric Power Applications, 14(13), 2616-2623.

(7) Sangirardi, M., Altomare, V., De Santis, S., & de Felice, G. (2022). Detecting damage

evolution of masonry structures through computer-vision-based monitoring methods.

Buildings, 12(6), 831.

(8) Liu, G., Shi, H., Kiani, A., Khreishah, A., Lee, J., Ansari, N., ... & Yousef, M. M. (2021).

Smart traffic monitoring system using computer vision and edge computing. IEEE

Transactions on Intelligent Transportation Systems, 23(8), 12027-12038.

(9) Wu, Z., Chen, Y., Zhao, B., Kang, X., & Ding, Y. (2021). Review of weed detection

methods based on computer vision. Sensors, 21(11), 3647.

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

103

(10) Li, Z. , Zhu, J. , Arumugam, K. , Bhola, J. , & Neware, R. . (2022). Research on computer

static software defect detection system based on big data technology. Journal of

Intelligent Systems, 31, 1055 - 1064.

(11) Lee, H., Hong, J., Wendimagegn, T. W., & Lee, H. (2021). Rail corrugation detection and

characterization using computer vision. Sensors, 21(24), 8335.

(12) Qiao, W., Ma, B., Liu, Q., Wu, X., & Li, G. (2021). Computer vision-based bridge damage

detection using deep convolutional networks with expectation maximum attention

module. Sensors, 21(3), 824.

(13) Shi, H., Zheng, L., Sun, S., Zhang, L., & Wang, L. (2020). Research on wire rope wear

detection based on computer vision. The Journal of Engineering, 2020(13), 517-519.

(14) Ljubovic, V., & Pajic, E. (2020). Plagiarism detection in computer programming using

feature extraction from ultra-fine-grained repositories. IEEE Access, 8, 96505-96514.

(15) Roy, S. D., Das, S., Kar, D., Schwenker, F., & Sarkar, R. (2021). Computer aided breast

cancer detection using ensembling of texture and statistical image features. Sensors,

21(11), 3628.

(16) Huang, H., Huang, T., Li, Z., Lyu, S., & Hong, T. (2021). Design of citrus fruit detection

system based on mobile platform and edge computer device. Sensors, 22(1), 59.

(17) Zhang, Y., Guo, Q., Zhang, Y., & Zhang, C. (2020). Fast and robust superpixel generation

method. IET Image Processing, 14(17), 4543-4553.

(18) Lecca, M., Torresani, A., & Remondino, F. (2020). Comprehensive evaluation of image

enhancement for unsupervised image description and matching. IET Image Processing,

14(16), 4329-4339.

(19) Nassu, B. T., Marchesi, B., Wagner, R., Gomes, V. B., Zarnicinski, V., & Lippmann, L.

(2021). A computer vision system for monitoring disconnect switches in distribution

substations. IEEE Transactions on Power Delivery, 37(2), 833-841.

(20) Hou, B., Zhang, C., & Yang, S. (2020). Computer vision tool-setting system of numerical

control machine tool. Sensors, 20(18), 5302.

(21) Hsia, S. C. , Wang, S. H. , Cheng, W. C. , & Chang, C. Y. . (2021). Intelligent blowgun

game scoring recognition system based on computer vision. IEEE Access, PP(99), 1-1.

(22) Duffy, T., Zachary, F., Hill, C., Sharp, A., Turner, S., Williams, J., ... & O'Brien, H. (2021).

The Effect of Image Stack Resampling on Manual Segmentation of the Putamen from

Magnetic Resonance Imaging. The FASEB Journal, 35.

(23) Feng, S., Hu, K., Fan, E., Zhao, L., & Wu, C. (2021). Kalman filter for spatial-temporal

regularized correlation filters. IEEE Transactions on Image Processing, 30, 3263-3278.

https://doi.org/10.17993/3ctic.2024.131.76-94

3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529

Ed.44 | Iss.13 | N.1 January - March 2024

104