Tomato Disease Identification Application Based on EfficientNetV2

Abstract

In each growth stage of tomato, there are many kinds of diseases, and artificial identification of crop diseases is easily interfered by subjective and environmental factors. Accurate identification of tomato diseases requires the guidance of agricultural experts, which consumes a lot of manpower and material resources. Some agricultural pest identification software exists on the market at present, but their identification accuracy is somewhat unstable. In this paper, nine common tomato diseases were identified accurately, based on EfficientNetV2 transfer learning, the value of mAP can reach 0.98 and the value of lost can reach 0.061. Meanwhile, the reasoning model is deployed to the cloud server, and the model is called based on the Flask framework, so as to infer and identify the photos uploaded by the Android application and return the results. The actual application test and comparative analysis show that the recognition accuracy of the recognition system designed in this paper is significantly higher than that of the existing commercial application software of the same type.