R. C. Dharmik
Assistant Professor, Department of Information Technology Yeshwantrao Chavan College of
Engineering, Nagpur, Maharashtra, (India).
Sushilkumar Chavhan
Assistant Professor, Department of Information Technology Yeshwantrao Chavan College of
Engineering, Nagpur, Maharashtra, (India).
Shashank Gotarkar
Asst. Professor, Dept. of Community Medicine, Jawaharlal Nehru Medical College, Datta Meghe
Institute of Medical Sciences, Sawangi (M), Wardha, (India).
Arjun Pasoriya
Software Test Team Lead,Amdocs, (Cyprus).
Reception: 07/11/2022 Acceptance: 22/11/2022 Publication: 29/12/2022
Suggested citation:
Dharmik, R. C., Chavhan, S., Gotarkar, S., y Pasoriya, A. (2022). Rice quality analysis using image processing
and machine learning. 3C TIC. Cuadernos de desarrollo aplicados a las TIC, 11(2), 158-164. https://doi.org/
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
Object Detection and its analysis are used in various fields. Rice quality evaluation subtask in
Agricultural industries is not exception for object Detection. Manual identification using image
processing techniques, Machine Learning Techniques and Deep Learning is also used for the rice
quality analysis. Due to Feature identification challenge machine Learning and Deep Learning are in
the demand. As rice is mostly used agricultural product so it is important to have the proper analysis
of the crops. In this study we proposed the used of image processing method with the help of Machine
Learning model. Rice grain morphological characteristics are what define a grain’s quality analysis.
The suggested method can operate efficiently with little expense.
Object Detection, Machine Learning, Morphological characteristics, Deep Learning.
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
The oldest and largest sector of the global economy is agriculture. Traditionally, a human sensory
panel uses physical and chemical features of food products to determine their quality (Mahale and
Korde, 2014)(Shah, Jain, and Maheshwari, 2013). In Asian nations, rice is a popular and widely eaten
cereal grain. It is widely accessible anywhere in the world. When rice is used for human consumption,
several items with value added are created. Quality is a major factor in the milled rice business. With
the growth of the import and export industries, quality assessment becomes increasingly crucial. The
dispensable items found in rice samples include paddy, chaff, broken grains, weed seeds, stones, etc.
These impurity levels affect the quality of the rice. As a complex problem it is solved by using image
processing techniques. There have been major advancements in the essential and cutting-edge
technology field of image processing such as canny edge detection algorithm (Mahale and Korde,
2014), Artificial Neural Network(Hamzah and Mohamed, 2020).
The approach of image processing is intended to preserve the integrity of the specifications. Image
manipulation involves applying certain procedures to a target image in order to produce a better and
more appealing image. And extract some useful data from the supplied image. Genetic algorithm
based LS-SVM(Chen, Ke, Wang, Xu, and Chen, 2012) was used which provides good result but
required lots of processing and complex operation. Later machine Learning based algorithms are used
for this classification and Analysis. ANN(Chen et al., 2012), SVM(Philip and Anita, 2017) is mostly
used algorithm which provides the good quality of results.
The main purpose of the proposed method is, to offer an alternative way for quality control and
analysis which reduce the required effort, cost and time by using other Machine Learning algorithms
and with object Detection. As rice quality analysis is control the diet and business of agriculture
industry, proper analysis is required. Image Processing Techniques and Machine Learning are tested
for the analysis. In this work we apply the object detection machine learning algorithm Region-based
Convolutional Neural Networks (R-CNN) with dimension reduction techniques. Results depict that
the results are less difference in the above methods.
This paper is organized as follows, and Section II describes the work of various researchers on Rice
quality analysis or detection. Section III gives a detailed overview of the technology used to select
good Rice and analyse it. Section IV uses this method to elaborate on the results. Section V provides
an overview of the results description, better performance than other results, and shortcomings.
Philip and Anita (2017), proposed new characteristics for rice grain categorization. For the
categorization of nine types of commercially accessible grains in the South Indian area, both spatial
and frequency-based criteria were applied. The classification is carried out in two phases, with the first
stage utilising the NB Tree classifier and the second stage utilising the SMO classifier.
Authors archives remarkable accuracy to the spatial features and suggested two work on real time.
Images. Parveen, Alam, and Shakir (2017) proposed image processing algorithm based some
characteristics with colour images. Characteristics wised results are obtained to user. Author suggested
applying the same with large data with more feature or characteristics. Kuchekar and Yerigeri (2018)
attempted to grade rice grains using image processing and morphological methods. Segmenting the
individual grains comes first, followed by pre-processing of the picture.
The grain’s geometrical characteristics, such as its area and the lengths of its main and minor axes, are
extracted and classified. The results have been determined to be positive. Rice is graded according to
the length of the grain. As a future scope it can be expanded in the future by focusing on moving
images and identifying additional qualities of rice grains. Kong, Fang, Wu, Gong, Zhu, Liu, and Peng
(2019) suggested to use an automated approach for extracting rice thickness based on edge properties.
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
The solution addressed the issue brought on by the structural similarity of edges by matching the
appropriate edge points based on the form of the edge rather than its texture. As a future scope authors
suggested to go with edge features extraction, with epipolar geometry to match the corresponding
points on the rice edge.
Avudaiappan and Sangamithra () analyse the visual features with image processing and MLP.
Authors used the SVM and Na¨ıve bays algorithm with 90% accuracy. As a future scope authors
focused on Non-Uniform Illumination with Transformation using top-hat so that rice can be classify in
long, normal or small category. Using a k-NN classifier, Wah, San, and Hlaing (2018) suggested an
image processing method and assessed three classes (30 images for each class).
Other studies Xiaopeng and Yong (2011), Yao, Chen, Guan, Sun, and Zhu (2009), Tahir, Hussin,
Htike, and Naing (2015) put more focus on identifying the grain’s apparent chalkiness. A grain with a
partly opaque or milky white kernel is said to be chalky. One of the key markers in the evaluation
process is the level of chalkiness. High levels of chalkiness in rice grains make them more likely to
shatter during milling, which will alter how they taste. An automated system for grading milled rice is
suggested by Wyawahare, Kulkarni, Dixit, and Marathe (2020).
Broken rice is an essential factor in rice grading. This technique may be used to determine the
percentage of broken rice from a sample’s picture. The relevant characteristics are retrieved from the
coloured pictures of the samples using particular preprogrammed processing procedures, and the
regression model is created. Estimating the percentage of broken rice requires less runtime than other
approaches since basic regression models are used. Lin, Li, Chen, and He (2018) offered a comparison
of two approaches—CNN and conventional methods—to identify rice grains with three distinct forms
(medium, round, and long grain). 5,554 photos were examined for calibration, and 1,845 images were
examined for validation. In the CNN approach, the experiments changed training parameters like
batch size and epochs. In a separate trial, they used conventional statistical techniques, and the
categorization accuracy they obtained ranged from 89 to 92%. As opposed to the conventional
approaches, the experiment employing the CNN method obtained a classification accuracy of 95.5%.
Benefits from the interplay between CNN and hyperspectral imaging were employed by
Chatnuntawech, Tantisantisom, Khanchaitit, Boonkoom, Bilgi¸c, and Chuangsuwanich (2018). Their
research used two sets of data, 414 samples from paddy rice and 232 samples from six different types
of milled rice. The accuracy of the suggested approach was 86.3%. In contrast, the SVM method on
the paddy seed dataset produced a result of 79.8%, whilst the accuracy of the other set was somewhat
off. By combining three machine learning techniques—kNN, SVM, and CNN—with hyper spectral
imagery, Qiu, Chen, Zhao, Zhu, He, and Zhang (2018) were able to identify four different types of rice
seeds. Two distinct spectral ranges were used in the experiment, and there were various numbers of
training samples.
In various studies, a hyperspectral camera was used to address the issue of classifying rice types. The
gadget, however, was expensive and complicated. Additionally, a quick computer, sensitive detectors,
and ample data storage were needed.
By observing the various literature we Identified architecture. Process is as follows.
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022
Fig 1: Architecture of the proposed system.
1. Training the data: To train the data Rice seeds image scanning needed then it Seed area segmentation is
done then Seed orientation was performed then proper data frame of images is form for training of data.
2. Training of data Training is done for both RCNN and statistical classical model.
3. Feature Extraction: For better accuracy we have to find the Features of the data frames. We used Greedy
Filter method to get the proper features.
4. Again training was performing and tests the accuracy on both RCNN and statistical classical model.
5. Dimension Reduction: to speed the process at each epoch we reduce the dimension by using PCA.
The below table gives average aspect ratio and classification which is based on kind of rice grain used. It
will show the exact value of parameters for the rice grain used in a bar graph where x-axis belongs to
particles and y-axis is average aspect ratio of parameters. Following Figure shows the ranges which
generally known as classification of rice grains that have been identified and training was done.
Table 1: classification of rice grains.
As per Architecture we have applied RCCN object detection method for identification of edges of the rice so that
classification can be done by training machine learning model. Here we train it using Recurrent neural network
(RNN) with activation function as Relu in hidden layers which can be helpful for RCNN. The Rice quality
analysis of application b uisng this application is shown below.
3C TIC. Cuadernos de desarrollo aplicados a las TIC. ISSN: 2254-6529
Ed. 41 Vol. 11 N.º 2 August - December 2022