Research On Variety Identification Of Rice Seed Using Computer Vision

In China, rice is one of chief crops in grain production. The quality of seed is an important factor influencing rice yield. The variety identification of rice seed is the most difficult in many quality factors. At present, the identification of rice seed variety mainly depends on chemical method and paddy field method. The two methods can give relatively more exact results but have many limitations. Application of the chemical method is hampered by the limited amount of sample and very high expense for inspection. The cycle of inspection using the paddy field method is too long to satisfy the demand of seeds circulation. This dissertation concentrated on research of variety identification for rice seed using computer vision.The main results of the research are as follows:1. A fiber circular halogen lamp with cold light source whose color temperature is 3300K, a lens of 50mm focal length, a 25mm extension tube and the white background were used in the image acquisition system of this research.2. Image preprocessing methods for variety identification of rice seed were studied. The grayscale threshold method was selected to background segmentation, which can reserve the color information of rice seed and remove its background. In this investigation, it was found that the blue proportion was very different between the background and the object. The setting of its threshold was based on the theory of minimum error probability. The weight-average method was applied to make the original image to the grayscale image. The grayscale threshold method was used in the image binary progress. The noise of the binary image was removed using the open and close operation of the morphological operation.3. Five color features were extracted from the image after background segmentation, including red average, green average, blue average, hue average and saturation average. Seven morphological features were extracted from the binary image using Blob analysis, including length, width, the ratio of length to width, area, parameter, roundness and radius of the inscribed cycle.4. For the various typical situations of rice seed variety recognition, the single-feature threshold method, the multi-feature threshold method and the artificial neural network were used in this investigation. The single-feature threshold method was used to recognize two variety rice seed for three different situations. For the recognition between xs11 and xy9308, either feature of length, the ratio of length to width, parameter, and roundness can give a perfect identification accuracies with 100%, 100%, 95%, 100%, which the thresholds were 450, 3.5, 1200, 1.8 respectively. For the recognition between xy5968 and xy9308, either feature of blue average, hue average and saturation average can give a perfect identification accuracy with 96%, 92%, 97%, using which the thresholds were 226, 236, 35respectively. For the recognition between ey7954 and syz3, the identification accuracy of 86% was achieved using the feature of radius of inscribed cycle that the threshold was set to 1.5. For the four variety recognition of ey7954, syz3xsl 1, z903, the multi-feature threshold method was used. The length and radius of inscribed cycle were selected and the identification accuracy was 90%, 80%, 92%, 100% respectively. For the four variety recognition of ey7954, syz3xsll, xy5968, z903, the three-layer BP neural network was used, which the identification accuracy was80. 3%, 73.5%, 85.4%, 77.6%, 7 5.0%respectively.5. A set of special software for variety identification of rice seed was developed.