Potato Plant Image Detection Based on Deep Learning

Abstract

Abstract. Potato is one of the most important food crops in the world. The information which extraction from high resolution remote sensing image is a new way to study the potato planting distribution and growth condition. For remote sensing target detection, a lot of people were used AdaBoost algorithm, SIFT algorithm, Tamura texture feature algorithm in the past. But it's just a feature of artificial extraction. Deep learning provides an effective framework for automatic extraction of target features. The experiment uses a simple but useful deep learning method (PCANet). After image segmentation, gray, binaryzation and filtering, the 42*48 of the potato plant images are trained and tested by feature extraction. The results showed that the detection rate of potato plants could reach 82.20%, the false detection rate was 12.66%, and the detection speed is 1.22-1.31 image per second, which could be applied to high efficiency fertilization, weeding and insect pests in order to achieve the purpose of increasing potato yield.