Akademik Veri Yönetim Sistemi
9th International Symposium on Multidisciplinary Studies and Innovative Technologies, ISMSIT 2025, Ankara, Türkiye, 14 - 16 Kasım 2025, (Tam Metin Bildiri)
This paper addresses the critical challenge of automated and precise plant species identification, a task fundamental to ecological monitoring, agricultural management, and biodiversity conservation. Conventional methodologies, often reliant on laborious manual morphological analysis or computationally expensive handcrafted feature engineering, frequently encounter limitations in scalability, inter-observer variability, and robustness against intra-species variations and environmental noise. To solve these inherent challenges, we propose LeafNet, a custom deep Convolutional Neural Network (CNN) architecture specifically designed for highly accurate and computationally efficient leaf image classification. LeafNet distinguishes itself through its remarkable low number of parameters (approximately 15K-35K parameters, depending on configuration), achieved by integrating components such as Ghost Modules for efficient convolutions, Shuffle Blocks for enhanced information flow, and Adaptive Activation functions for improved representational capacity. Our model capitalizes on the inherent hierarchical feature learning capabilities of CNNs to autonomously extract salient and discriminative visual patterns directly from raw leaf image data, thereby circumventing the need for explicit feature extraction. The efficacy of the proposed LeafNet was rigorously evaluated through extensive experimentation on the PlantVillage dataset, yielding a notable classification accuracy of %98.34 and an average AUC of 0.9998. These compelling experimental results underscore the superior predictive performance, enhanced generalization capability, and inherent robustness of our neural network-based paradigm, positioning it as a highly promising and scalable solution for advanced automated plant identification, early detection of phytopathological conditions, and systematic biodiversity assessment, particularly in resource-constrained environments.