Segmentation of rice disease spots based on improved BPNN

An adaptive learning rate Backpropagation Neural Network (BPNN) is proposed to image segmentation of rice disease spots. Rice blast is a common disease of rice and is tested in this paper. Firstly, the combination of different color feature parameters is selected as the input of the BPNN. Secondly, a BPNN with 5 input, 10 hidden neurons and 1 output is constructed to rice blast spots segmentation. The experimental results show that the converging speed of training the improved BPNN is faster than that of the standard BPNN. To test the accuracy of the algorithm, manually segmented images are compared with those segmented automatically. Results show that the highest accuracy of 99.8% is found in the test, and the mean segmentation accuracy rates and the mean misclassification rates of two methods are very close. This method can be applied to segment the other rice disease spots and the disease spots of other crops.