Involution-Infused DenseNet with Two-Step Compression for Resource-Efficient Plant Disease Classification (2506.00735v1)

Abstract: Agriculture is vital for global food security, but crops are vulnerable to diseases that impact yield and quality. While Convolutional Neural Networks (CNNs) accurately classify plant diseases using leaf images, their high computational demands hinder their deployment in resource-constrained settings such as smartphones, edge devices, and real-time monitoring systems. This study proposes a two-step model compression approach integrating Weight Pruning and Knowledge Distillation, along with the hybridization of DenseNet with Involutional Layers. Pruning reduces model size and computational load, while distillation improves the smaller student models performance by transferring knowledge from a larger teacher network. The hybridization enhances the models ability to capture spatial features efficiently. These compressed models are suitable for real-time applications, promoting precision agriculture through rapid disease identification and crop management. The results demonstrate ResNet50s superior performance post-compression, achieving 99.55% and 98.99% accuracy on the PlantVillage and PaddyLeaf datasets, respectively. The DenseNet-based model, optimized for efficiency, recorded 99.21% and 93.96% accuracy with a minimal parameter count. Furthermore, the hybrid model achieved 98.87% and 97.10% accuracy, supporting the practical deployment of energy-efficient devices for timely disease intervention and sustainable farming practices.

• The paper proposes a hybrid Involution-infused DenseNet model with two-step compression using weight pruning and knowledge distillation.
• The compressed models achieve high accuracy (up to 99.21%) on datasets like PlantVillage while significantly reducing parameters and computational cost.
• This method facilitates the practical deployment of CNNs for real-time plant disease identification on resource-constrained mobile and edge devices in agriculture.

Involution-Infused DenseNet with Two-Step Compression for Resource-Efficient Plant Disease Classification

The paper "Involution-Infused DenseNet with Two-Step Compression for Resource-Efficient Plant Disease Classification" by Ahmed et al. addresses the computational challenges faced by Convolutional Neural Networks (CNNs) in predicting plant diseases and proposes a novel hybrid model with a two-step compression approach. The researchers aim to enhance the practical utility of CNNs for plant disease identification, especially in resource-constrained environments like mobile devices for real-time agricultural monitoring.

The core of the paper lies in the integration of innovative techniques—Weight Pruning and Knowledge Distillation—coupled with the application of Involutional Layers to a DenseNet architecture. The weight pruning technique systematically reduces non-essential parameters to streamline computational demands, while knowledge distillation transfers knowledge from larger, complex models to smaller, more efficient student models without significant losses in performance. This hybridization involves the combination of DenseNet architecture with Involutional Layers, which capture spatial features more effectively than standard convolutions, thereby enhancing feature extraction.

The paper makes a compelling case for its two-step compression framework. Initial results demonstrate the efficacy of models like ResNet50, which achieved 99.55% accuracy on the PlantVillage dataset, validating the robustness of the approach. Moreover, the DenseNet-based compressed models showed commendable accuracy (up to 99.21%) with significantly reduced parameters, highlighting improved computational efficiency. The hybrid model further underscores the effectiveness of this method by achieving 98.87% accuracy on the PaddyLeaf dataset, supporting its application in energy-efficient devices.

The implications of this research extend significantly into precision agriculture. By presenting a viable method to adapt CNNs for real-time deployment on edge devices, the paper advances methodologies for rapid disease identification, which can lead to sustainable farming practices and enhanced crop management. This research aligns with global food security efforts by providing farmers with tools to optimize yield and quality output while utilizing technological advancements effectively.

Looking towards future trends, the paper elucidates potential paths in AI-driven agriculture. By reducing the computational burden of deep learning models, the authors suggest possibilities in extending these applications to real-world scenarios, such as drones and other low-power devices, to facilitate more comprehensive crop monitoring systems. Additionally, the approach could benefit other fields requiring efficient AI models under resource constraints, like medical image analysis or autonomous vehicle systems.

In conclusion, Ahmed et al.'s paper is an insightful contribution to the domain of AI in agriculture, offering a viable solution to a critical bottleneck in deploying advanced CNN architectures in resource-limited settings. Through this thoughtful integration of compression techniques and hybridization, the paper stands as a testament to leveraging AI's potential in enhancing agricultural productivity and food security.