Recognition of pests on crops with a random subspace classifier

The purpose of this study is to develop and test a recognition system for the Colorado potato beetle. This task is very important for localizing the beetles and reducing the pesticide volume used to protect the harvest. We employ a beetle image dataset that contains 25 images representing different beetle positions and varying numbers of beetles. These images were collected from the Internet. Our recognition system is based on a special neural network, the random subspace classifier (RSC). We calculate the brightness, contrast, and contour orientation histograms of the images and use them as features and inputs to the RSC neural classifier. In addition, we describe the RSC structure and algorithms and analyse the obtained results. We obtained the best recognition rate of 85%.