Authors:
R. P. Pranav, Balika J. Chelliah, Kawsher Rahman
Addresses:
Department of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of General Medicine, Beanibazar Cancer and General Hospital, Sylhet, Bangladesh.
Abstract:
There is a risk that plant leaf diseases will affect crop yields all over the world; it is necessary to make a diagnosis as quickly and accurately as possible. Even though human errors slow down the system and produce outcomes that are not systematic, the system is driven by expert insights. The advocated method uses Explainable Artificial Intelligence (XAI) to automatically identify potato leaf illnesses caused by bacteria, fungi, nematodes, viruses, pests, Phytophthora, and healthy potato leaves. The purpose of this investigation is to examine the similarities and differences between the accomplishments of ResNet50 and InceptionV3 and those of Star3-Net, which is an innovative architecture. The ResNet50 and VGG16 weights, in addition to the InceptionV3 weights, are included in Star3-6 via transfer learning. According to the studies, Star3-Net performs better than both ResNet50 (68.1%) and InceptionV3 (70.2%), achieving a maximum classification accuracy of 75.4%. The process of transfer learning helps enhance model categorisation by combining the attributes of several models. Using Local Interpretable Model-Agnostic Explanations (LIME) and Star3-Net, it is feasible to visually diagnose disease spots on plant leaves.
Keywords: Plant Disease Classification; Explainable AI; Deep Learning; Agricultural Disease Detection; Crop Production; Model Interpretability; Infections and Viruses.
Received: 29/11/2024, Revised: 18/03/2025, Accepted: 18/06/2025, Published: 09/12/2025
DOI: 10.64091/ATIHL.2025.000207
AVE Trends in Intelligent Health Letters, 2025 Vol. 2 No. 4 , Pages: 198-207