Authors:
Swathi Baswaraju, K. A. Jayasheel Kumar, Thirumalraj Karthikeyan, S. Venkatasubramanian, Sheila Agnes Vidot
Addresses:
Department of Computer Science and Engineering, New Horizon College of Engineering, Bengaluru, Karnataka, India. Department of Automobile Engineering, New Horizon College of Engineering, Bengaluru, Karnataka, India. Department of Artificial Intelligence, Trichy Research Labs, Quest Technologies, Tiruchirappalli, Tamil Nadu, India. Department of Computer Science and Business Systems, Saranathan College of Engineering, Tiruchirappalli, Tamil Nadu, India. Department of Medical Consultant, IPS Health, Mahé, Victoria, Seychelles.
Abstract:
Automatic plant disease detection is essential because it minimizes the laborious task of keeping an eye on large farms and identifies diseases early on, when they can still be prevented from causing further damage to plants. This scenario has a significant impact on a nation's economy and harms plant health by reducing production. The suggested model used Deep Learning (DL) to achieve the best possible classification accuracy for leaf diseases. For additional processing, the suggested model used the Plant-Microbe Biology and Plant Pathology sections of the Cornell University Dataset. As a crucial component of image preprocessing, denoising can significantly improve image quality, benefiting subsequent operations such as feature extraction and image segmentation. This study used the Non-Local Mean (NLMM) filtering technique for image denoising. Following preprocessing, Convolutional neural networks, encoder-decoders, and Swin Transformers were the three branches into which the extraction process was integrated, creating the SwinTransConv-ED model. Following the extraction procedure, a novel leaf image classification framework, Hybrid-Transformer and CNN (HTConvNet), built on the Transformer architecture, is proposed. In this study, the hyperparameter tuning procedure is used to attain the highest level of accuracy. An improved teaching-learning-based optimisation (ETLBO) to tune the HTConvNet classifier's hyperparameters.
Keywords: Deep Learning; Image Denoising; Swin Transformer; Leaf Image Classification; Teaching and Learning; Non-Local Moment Mean; Hybrid-Transformer and CNN; Swintransconv-ED.
Received: 30/10/2024, Revised: 16/02/2025, Accepted: 04/05/2025, Published: 07/09/2025
DOI: 10.64091/ATIHL.2025.000174
AVE Trends in Intelligent Health Letters, 2025 Vol. 2 No. 3 , Pages: 163-177