Challenges and Trends in Optimized CNN for Leaf Feature Extraction Optimization in Multi-Disease Plant Detection

Abstract

Early detection of plant diseases is crucial for ensuring crop health and preventing yield losses. Convolutional Neural Networks (CNN) have experienced rapid development in plant disease image recognition due to their ability to extract significant visual features from plant leaves. However, optimal results require CNN architecture customization according to unique disease and crop characteristics. While this approach offers high accuracy and efficiency, various challenges hinder widespread application, including limited representative datasets, high computational requirements, and difficulties in designing generalizable models for different field scenarios. Additionally, model interpretability issues often arise, hindering large-scale adoption among agricultural practitioners. This systematic literature review addresses these challenges and explores recent trends in optimized CNN development for plant leaf feature extraction. Through PRISMA methodology, 26 peer-reviewed studies from 2018-2024 were analyzed from Scopus Q1-Q4 journals. Key findings include the effectiveness of data augmentation techniques (improving dataset diversity by 40-60%), transfer learning approaches (reducing training time by 50-70%), and hybrid model integration (achieving 85-95% accuracy rates). Architecture improvements and optimization algorithms help overcome computational constraints, with lightweight models reducing processing time by 30-50% while maintaining 90%+ accuracy. This study provides comprehensive guidance for researchers and practitioners in developing more adaptive, accurate, and efficient plant disease detection solutions, ultimately improving agricultural yields and global food security.