A Deep-Learning-Based Low-Altitude Remote Sensing Algorithm for Weed Classification in Ecological Irrigation Area

Abstract

With the development of ecological irrigation area at present, it requires higher detection and control of weeds in irrigation area. In this paper, aiming at the ecological irrigation area, a classification method of weeds based on convolutional neural network (CNN) is proposed. By collecting 3 kinds of weeds and 3 kinds of crops as data sets, through cutting, rotating and so on, data is transported to the CNN. Finally, 6 categories of classifications are implemented. By using the pre-trained AlexNet network for transfer learning, single CPU, single GPU, and double GPUs training experiments are performed in matlab2018(a). The classification results show that the recognition rate of weeds can reach 99.89%. In order to prevent and control specific weeds, a method of detecting single weeds density is also presented in this paper. The accurate monitoring of weeds in irrigation area can be realized through the method proposed in this paper, and there is basis for precise weed control in later stage.