Multiple Fruits Plant Leaf Classification Using Deep Learning Models

Abstract

Leaf diseases present substantial challenges to global agricultural productivity, causing significant economic losses for farmers. Addressing these issues necessitates effective measures for disease identification. This thesis explores advanced technological solutions, specifically the integration of machine learning and computer vision, to establish intelligent farming through early detection of leaf diseases. An essential tool in aiding farmers is an intelligent analyzer capable of autonomously discerning various leaf diseases. This research encompasses the investigation, conceptualization, and implementation of such an analyzer, utilizing cutting-edge computer vision and machine learning techniques for disease identification based on leaf appearance. The study involves a diverse range of experiments and assessments, exploring various segmentation, feature extraction, and classification methods to pinpoint the most efficacious approach. Beyond traditional machine learning, the research delves into the capabilities of deep learning, leveraging its potential for enhanced accuracy and real-time analysis. The primary focus is on developing a robust and adaptable system that can accurately identify common leaf diseases. The target audience for this technology extends to users seeking a rapid and complimentary diagnosis of common leaf diseases, accessible at any time of the day. Emphasizing the potential for real-time, deep learning-based solutions, the thesis

V underscores the significance of such technology in precision agriculture. The findings contribute to the broader field of agricultural technology, offering insights into the practical application of deep learning for sustainable and efficient farming practices. This study presents 14 distinct plant species from the dataset, employing compact (CNNs) and RestNet with “transfer learning”. The models are meticulously trained on plant leaf data,incorporating various data augmentation techniques. Notably, the integration of data augmentation demonstrates a remarkable enhancement in the classification accuracy. Furthermore, proposed models extend their applicability f o the classifications of 38 disease classes within the dataset. The developed RestNet model exhibits an outstanding classification accuracy of 99.2%, showcasing its robustness in plant species identification. In contrast, the VGG model, while delivering a classification accuracy of 30.5%, reveals insights into its comparative performance. The findings underscore the efficacy of the proposed models in advancing the state-of-the-art in plant species and disease classification. This work contributes to the growing body of knowledge in the field of computer vision applied to agriculture,