An Application for Identification of Malignant Weeds in Cereal Fields Based on Neural Network

In the process of agricultural production, the existence of malignant weeds seriously affects the yield and quality of crops, and especially the malignant weeds represented by Johnson grass, Aegilops tauschii and Nut grass, which are not only extremely similar to crops in morphology, but also have the characteristics of fierce growth and high reproduction rate. In this paper, we designed an image recognition and processing method, especially focusing on the automatic identification and determination of these weeds, in order to solve the defects of low efficiency and error when using manual identification of weeds, and improve the efficiency of weed control in the crop production cycle. Based on deep learning techniques, this study trained the model by means of transfer learning. At the same time, the steps of data processing, such as data collection, data cleaning and purification, as well as the details of model training are listed. On this basis, an application was developed to identify weed species by taking and selecting pictures. Through using cloud computing to improve performance, and to speed up the system, the application can almost achieve 95% identification accuracy.