A Benchmark of Deep Learning Models for Multi-leaf Diseases for Edge Devices

Abstract

Every season, leaf diseases are one of the main causes affecting the production of many crops, which cause enormous damage to farmers. To minimize the loss, deep learning techniques are utilized to detect leaf infection and have wildly outperformed the traditional method of manual detection. However, deploying such models is a challenge since devices in the field normally have limited resources and low computational power while large datasets have to be used. Therefore, in this paper, we benchmark the most popular deep learning models for multi-leaf disease detection to gauge which model is the most suitable for real deployment. Using a real-world large-scale dataset from PlantVillage and a Raspberry Pi 3, we found that MobileNet V3 provides a reliable accuracy of 96.58%, small Inference/Initialization time of 127 ms and 11 ms respectively, requires only 7.4 MB of memory in total, and hence the most appropriate choice for a real farm.