Toledic: Tomato Leaf Disease Classification Lite

Tomatoes are edible berries with the rich umami flavor. Since, identification of diseases in tomato leaves is essential in Indian market to maximize the crop yield and minimize the crop loss. As such, this paper investigates the usage of lightweight Convolution Neural Networks deployed on a Raspberry Pi 4 for real-time diagnosis of tomato leaf diseases. In this paper, I focus on computational constraint on the Raspberry Pi 4 at the cost of accuracy and introduce a novel model, Toledic-Lite. We evaluate and compare various conventional lightweight models based on ResNet 50, EfficientNet v4, DenseNet 121, ConvNext v2_Nano and VGG16 metrics such as accuracy, inference time and memory usage, and identify conventional edge computation performance. Toledic-Lite’s functionality is meant to bring the data processing process in the complex computational tasks of Arm based processors into more efficient and accurate domain by optimizing the accuracy and the result as compared to the conventional model. The objective is to develop a cost effective, efficient and accurate solution for deployment on farms. The work presents models using multiple architectures on the Raspberry Pi platform that balance accuracy and computational efficiency. Overall, it is demonstrated that lightweight CNNs can accurately classify tomato leaf diseases to support the development of intelligent and sustainable agricultural practices. This work demonstrates a significant contribution to the exploration of the application of edge computing for disease management in agriculture, focusing on a need to intervene early to protect crop health and improve agricultural output.