A Lightweight Siamese Network Framework for Effective Tomato Disease Recognition

Food Security is a major global concern, and artificial intelligence (AI) is playing an increasingly crucial role in this area. In this context, automatic recognition of plant diseases, such as tomato disease detection from leaf images, is of particular importance to prevent crop losses and ensure sustainable food production. Deep learning-based tomato disease recognition methods have shown promising results, but they face major challenges. They often require large sets of labeled data, which can be expensive and time-consuming to obtain. Moreover, these traditional deep learning models consume a lot of memory and storage due to the large number of parameters they use. In this work a lightweight framework based on a Siamese network was developed for the automatic recognition of tomato leaf diseases. This lightweight framework achieved a remarkable accuracy on a subset of tomato data from the PlantVillage dataset. The experimental results demonstrate the effectiveness of this framework in handling unbalanced and small data. In addition, the backbone deep network integrated into this framework is distinguished by its lightness, which makes it far inferior to existing light deep networks.