Towards precision agriculture: A fine grained lightweight model for apple leaf disease detection

This is the authors' abstract. We don't add key points for this paper.

Apples are loved by people for their rich nutritional value. Apple leaf diseases significantly impact fruit quality and yield, necessitating precise disease classification. Currently, disease detection faces a number of challenges, such as the significant effects of environmental factors such as climate change, lighting conditions, and growth stages, as well as similar symptoms between different diseases and different manifestations of the same disease under different conditions, all of which increase the difficulty of identification. In this study, a lightweight, fine-grained apple leaf disease classification network (LM2-Net) is proposed to address these challenges. Firstly, we designed a lightweight enhanced spatial representation backbone(LESR), which enhances capabilities without adding complexity. Then, a multi-dimensional cosine fine-grained optimization algorithm (MDFO) was proposed to identify disease similarity problems in complex environments more accurately. In addition, a Mamba-guided edge dynamic enhancement algorithm (MGEA) was developed in this paper to effectively address the diverse symptom recognition challenges of the same disease in different geographical environments. Experiments on the Apple Leaf Disease Dataset (AppleLeaf9) demonstrated that our method achieved an average accuracy of 98.15% and an F1 score of 96.90%, confirming its effectiveness. In addition, we conducted on-site testing of mobile devices at an apple orchard in Qingyang City, Gansu Province, China, and the test results showed good performance. The implementation code and partial datasets are available at https://github.com/AroundSeeYou/LM2-Net . This work provides both theoretical advancements in fine-grained plant disease classification and practical solutions for precision agriculture applications.