Image Classification of Rice Diseases Based on Metric Learning

At present, crop diseases are still an important reason affecting the yield, safety and economy. With the continuous development of information technology and its wide application in agriculture, crop disease identification is moving from the initial artificial vision judgment to intelligence and digitalization. However, the computer vision method based on deep learning still depends on the large amount of labeled data. To solve this problem, a method MST-RelationNet based on metric learning is proposed. 500 images of six kinds of rice leaf diseases are labeled to form a dataset of rice leaf diseases, and a number of experiments are carried out on this basis. As can be seen from the results, the method that we proposed, has better performance than classic method ProtoNet, and the accuracy is improved by about $2 \%-5 \%$. The development of this method is prospected, which provides a basis for the application and research of computer vision technology in crop disease recognition.