Abstract.The rice brown planthopper (BPH) outbreak is one of several causes of damage to rice crops in Thailand. A traditional way to monitor the early outbreak is to routinely and randomly count the density of BPHs spreading around the rice field. This article presents an assistive tool to monitor the BPH by using automatic image processing. Smart phone devices with a sufficient camera quality are currently affordable and convenient for farmers to capture images from their rice fields. Based on the Support Vector Machines algorithm trained on color and Gray Level Co-occurrence Matrix (GLCM) image features, the proposed system not only automatically detects the position of BPHs in the collected images, but is also able to classify the life stage of each hopper. The use of a red-frame mark on the camera screen to guide BPH image capturing helps improving the overall processing accuracy. Field experiments with the Rice Department of the Ministry of Agriculture and Cooperatives of Thailand shows the proposed system achieved an approximately 89% detection F-measure and an 87% BPH life-stage classification accuracy. Moreover, this article illustrates the preciseness of BPH density prediction with respect to the different numbers of sampling images from the rice field. The result suggests farmers to take at least 40 images per 1,600 square meters in order to gain more than 87% prediction accuracy. Keywords: BPH classification, BPH life-stage, Brown planthopper (BPH), Image processing, Mobile device, Support vector machines, Rice.
On Combining Convolutional Autoencoders and Support Vector Machines for Fault Detection in Industrial Textures
