Rice Sheath Blight Recognition Based On Neural Networks

Food security has become a great challenge to the development of our country. Meanwhile, rice is the largest cereal crop in our country, and the production could apparently affect the food security situation. This study was conducted to recognize the rice sheath blight by neural networks, which could be useful in rural district where with low level of education, lack of technology and insufficient agricultural consulting. Additionally, neural networks could effectively promoting the recognition and inhibition on rice sheath blight, reduce the losses of famer and guarantee the food security.This study discussed the improvement of image analysis for rice sheath blight through image preprocessing, feature extraction and the arrangement of neural net parameters. The results and conclusion are demonstrated below.1. Estimate the preprocessing methods of rice sheath blight sample images, determined the applicable size of each sample. After that, we approached the sample images with the regular methods including flatting, enhancement and segmentation. We found that median filter was useful to increase the identification rate, while the enhancement of images could decrease the identification rate.2. We extracted the feature of the sample images and found that color and texture could apparently useful for recognizing rice sheath blight. While, shape was useless to promoting the identification of rice sheath blight.3. Experimental Results showed that the parameters of BP model could have an effect on the successful rate of image identification and the training time, and the best neural net for recognition was found by trained with the optimal parameters, and the recognition rate of rice sheath blight was 89.5%...