Determination of the Varieties of Rice Kernels Based on Machine Vision and Deep Learning Technology
01 Dec 2017-pp 169-172
TL;DR: A machine vision together with the developed neural network architectures could be used as a tool to achieve better and more objective rice quality evaluation at trading points within the rice marketing system, and it also provide a superior alternative for other research.
Abstract: In this paper, we present a model of a convolutional neural network for automatic extraction of several features of the rice kernels from a gray image The system developed convolutional neural network which is consisted of 7 layers and receives a gray image that measures 200 × 200 pixels as its input Therefore, there are 40000 neurons on the network at the input level The following layer is convolutional with the set of 6 filters Next is the subsampling layer with the maximumvalue function Then, we have one more convolutional layer of 12 filters and the subsampling layer with the maximumvalue function The final is the fully connected layer of 3 neurons The convolutional Neural Networks detector developed were able to identify the grain samples at overall average accuracies of 9952% The study results have demonstrated the capability and potential of machine vision with well-trained convolutional neural network detector for varietal types identification of rice grain samples With the comparably high accuracy of classification obtained, a machine vision together with the developed neural network architectures could be used as a tool to achieve better and more objective rice quality evaluation at trading points within the rice marketing system, and it also provide a superior alternative for other research
Citations
More filters
TL;DR: In this article , a machine vision system is developed to first construct a dataset of 8048 high-magnification (4.5 x) images of damaged rice refractions, that are obtained through the on-field collection.
28 citations
Journal Article•
TL;DR: In this article, a method was developed for the determination of the size and size distribution of rice and the amount of broken rice kernels using flatbed scanning (FBS) and image analysis (IA).
Abstract: A method was developed for the determination of the size and size distribution of rice and the amount of broken rice kernels using flatbed scanning (FBS) and image analysis (IA). The rice was placed on the glass plate of the scanner and covered with a black sheet of paper. A fully automatic procedure was developed using freeware IA software and standard spreadsheet software. The method was tested on parboiled and regular-milled white rice of different varieties and compared to manual analysis by weighing after visual separation of whole and broken kernels and by measuring the length and width of rice kernels using a sliding calliper. The FBS method is fast, easy to use and cheap. It yields the same accuracy and better precision than the more time-consuming manual method. Analysis by FBS-IA takes about 3 min per sample compared to about 30 min for manual analysis. The procedure requires a PC with standard desktop scanner.
9 citations
TL;DR: In this article , a machine vision unit was set up for image acquisition using an area scan camera for the Red, Green and Blue (RGB) and monochrome images of five cultivated rice varieties and a weedy rice seed variant.
Abstract: Weedy rice infestation has become a major problem in all rice-growing countries, especially in Malaysia. Challenges remain in finding a rapid technique to identify the weedy rice seeds that tend to pose similar taxonomic and physiological features as the cultivated rice seeds. This study presents image processing and machine learning techniques to classify weedy rice seed variants and cultivated rice seeds. A machine vision unit was set up for image acquisition using an area scan camera for the Red, Green and Blue (RGB) and monochrome images of five cultivated rice varieties and a weedy rice seed variant. Sixty-seven features from the RGB and monochrome images of the seed kernels were extracted from three primary parameters, namely morphology, colour and texture, and were used as the input for machine learning. Seven machine learning classifiers were used, and the classification performance was evaluated. Analyses of the best model were based on the overall performance measures, such as the sensitivity, specificity, accuracy and the average correct classification of the classifiers that best described the unbalanced dataset. Results showed that the best optimum model was developed by the RGB image using the logistic regression (LR) model that achieved 85.3% sensitivity, 99.5% specificity, 97.9% accuracy and 92.4% average correct classification utilising all the 67 features. In conclusion, this study has proved that the features extracted from the RGB images have higher sensitivity and accuracy in identifying the weedy rice seeds than the monochrome images by using image processing and a machine learning technique with the selected colour, morphological and textural features.
5 citations
01 Jan 2018
TL;DR: In this project, seed holder was designed to hold rice seed for image acquisition purposes, and four different colors such as black, blue, green and red was painted on the seed holder.
Abstract: One of the crucial part in the development of machine vision for rice seed identification are the design of the seed holder itself. In this project, seed holder was designed to hold rice seed for image acquisition purposes. Four different colors such as black, blue, green and red was painted on the seed holder. Effect of background colors on rice seeds image segmentation were tested under machine vision setup. Simple rice seed parameters such as seed length and width were measured using image processing technique programmed in LabVIEW software. Percentage error for each background color was calculated based on the actual legth and width of the rice seed. Blue background color was found to provide good contrast for estimation of length and width with accuracy less than 2% and 5%, respectively.
3 citations
Cites methods from "Determination of the Varieties of R..."
...[4] use convolutional neural network to automatically extract rice kernel features from grey image and claimed the technique average accuracies of 99....
[...]
TL;DR: In this article , an efficient desktop-application-based rice quality evaluation system based on computer vision and machine learning is presented, which is based on seven main features including grain length, width, weight, yellowness, broken, chalky, and/or damaged kernels for six different types of rice: IRRI-6, PK386, 1121 white and Selah, Super kernel basmati brown, and white rice.
Abstract: The assessment of food quality is of significant importance as it allows control over important features, such as ensuring adherence to food standards, longer shelf life, and consistency and quality of taste. Rice is the predominant dietary source of half the world’s population, and Pakistan contributes around 80% of the rice trade worldwide and is among the top three of the largest exporters. Hitherto, the rice industry has depended on antiquated methods of rice quality assessment through manual inspection, which is time consuming and prone to errors. In this study, an efficient desktop-application-based rice quality evaluation system, ‘National Grain Tech’, based on computer vision and machine learning, is presented. The analysis is based on seven main features, including grain length, width, weight, yellowness, broken, chalky, and/or damaged kernels for six different types of rice: IRRI-6, PK386, 1121 white and Selah, Super kernel basmati brown, and white rice. The system was tested in rice factories for 3 months and demonstrated 99% accuracy in determining the size, weight, color, and chalkiness of rice kernels. An accuracy of 98.8% was achieved for the classification of damaged and undamaged kernels, 98% for determining broken kernels, and 100% for paddy kernels. The results are significant because the developed system improves the local rice quality testing capacity through a faster, more accurate, and less expensive mechanism in comparison to previous research studies, which only evaluated four features of the singular rice type, rather than the seven features achieved in this study for six rice types.
2 citations
References
More filters
Proceedings Article•
03 Dec 2012TL;DR: The state-of-the-art performance of CNNs was achieved by Deep Convolutional Neural Networks (DCNNs) as discussed by the authors, which consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax.
Abstract: We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overriding in the fully-connected layers we employed a recently-developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.
73,978 citations
TL;DR: A fast, greedy algorithm is derived that can learn deep, directed belief networks one layer at a time, provided the top two layers form an undirected associative memory.
Abstract: We show how to use "complementary priors" to eliminate the explaining-away effects that make inference difficult in densely connected belief nets that have many hidden layers. Using complementary priors, we derive a fast, greedy algorithm that can learn deep, directed belief networks one layer at a time, provided the top two layers form an undirected associative memory. The fast, greedy algorithm is used to initialize a slower learning procedure that fine-tunes the weights using a contrastive version of the wake-sleep algorithm. After fine-tuning, a network with three hidden layers forms a very good generative model of the joint distribution of handwritten digit images and their labels. This generative model gives better digit classification than the best discriminative learning algorithms. The low-dimensional manifolds on which the digits lie are modeled by long ravines in the free-energy landscape of the top-level associative memory, and it is easy to explore these ravines by using the directed connections to display what the associative memory has in mind.
15,055 citations
"Determination of the Varieties of R..." refers background in this paper
...The larger the number of layers in a neural network, the more powerful the classifier produced and the more complex features can be extracted from data [11]....
[...]
07 Jun 2015
TL;DR: This paper proposes a simple convolutional net architecture that can be used even when the amount of learning data is limited and shows that by learning representations through the use of deep-convolutional neural networks, a significant increase in performance can be obtained on these tasks.
Abstract: Automatic age and gender classification has become relevant to an increasing amount of applications, particularly since the rise of social platforms and social media. Nevertheless, performance of existing methods on real-world images is still significantly lacking, especially when compared to the tremendous leaps in performance recently reported for the related task of face recognition. In this paper we show that by learning representations through the use of deep-convolutional neural networks (CNN), a significant increase in performance can be obtained on these tasks. To this end, we propose a simple convolutional net architecture that can be used even when the amount of learning data is limited. We evaluate our method on the recent Adience benchmark for age and gender estimation and show it to dramatically outperform current state-of-the-art methods.
1,046 citations
"Determination of the Varieties of R..." refers background in this paper
...A part of brain called neocortex is having layered architecture [10]....
[...]
TL;DR: The operation of tolerating positional error a little at a time at each stage, rather than all in one step, plays an important role in endowing the network with an ability to recognize even distorted patterns.
1,037 citations
"Determination of the Varieties of R..." refers background in this paper
...The neural network consists of several layers of neurons [12] and computes a multidimensional function of several variables....
[...]
Posted Content•
TL;DR: This work reports state-of-the-art and competitive results on all major pedestrian datasets with a convolutional network model that uses a few new twists, such as multi-stage features, connections that skip layers to integrate global shape information with local distinctive motif information.
Abstract: Pedestrian detection is a problem of considerable practical interest. Adding to the list of successful applications of deep learning methods to vision, we report state-of-the-art and competitive results on all major pedestrian datasets with a convolutional network model. The model uses a few new twists, such as multi-stage features, connections that skip layers to integrate global shape information with local distinctive motif information, and an unsupervised method based on convolutional sparse coding to pre-train the filters at each stage.
713 citations
Additional excerpts
...Recently, deeplearning neural networks have been applied to solve various problems of machine learning, such as image classification [8], pedestrian detection [9] and house number digit classification which obtained the best results....
[...]