Optimization of Plant Disease Detection and Classification Using an Antlion-Optimized VGG16 Model with Fuzzy Rough C-Means Segmentation
Abstract
The agricultural sector in India, supporting over 65% of the population, faces significant challenges from plant diseases that threaten crop productivity and food security. Traditional disease identification methods are often slow and require expert knowledge. This paper proposes a novel, automated framework for accurate plant disease detection by integrating advanced image processing with deep learning. The methodology employs a Median filter for image pre-processing, the Fuzzy Rough C-Means (FRCM) clustering algorithm for robust segmentation of diseased leaf regions, and a Convolutional Neural Network (CNN) for classification. The core innovation lies in enhancing a standard VGG16 CNN architecture using the Antlion Optimization (ALO) algorithm to optimize its hyperparameters, specifically the number of neurons in a fully connected layer, thereby improving feature learning and classification performance. Trained and tested on a dataset of cotton leaf images encompassing healthy samples and four disease types, the proposed ALO-enhanced VGG16 model achieved a high average classification accuracy of 93.33%. This performance surpasses that of other standard classifiers, including basic CNN, SVM, and ResNet models. The findings demonstrate that the integration of metaheuristic optimization with deep learning offers a powerful, scalable tool for precise plant disease diagnosis, with the potential to aid sustainable agricultural practices.
Keywords
Download Options
Introduction
Agriculture is a vital global industry and a primary source of livelihood. Plant diseases, caused by pathogens such as fungi, bacteria, and viruses, represent a major threat to crop yield and quality, leading to significant economic losses [1, 2]. Early and accurate detection is crucial for implementing effective management strategies. Conventional visual inspection by experts is subjective, time-consuming, and not scalable. Consequently, there is a growing need for automated, reliable, and rapid diagnostic systems.
Computer vision and artificial intelligence offer promising solutions. Initial approaches utilized basic image processing for colour and texture analysis [6], while machine learning models like Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs) were applied for classification [7, 8]. The advent of deep learning, particularly Convolutional Neural Networks (CNNs), has revolutionized the field due to their superior ability to automatically learn hierarchical and discriminative features from raw image data [9]. Models like VGG16 have become benchmarks in image classification tasks. Despite their success, standard CNNs may not be optimally configured for specific domains like plant pathology. Their fixed architectures might be suboptimal for learning the distinctive features of various leaf diseases. This creates a research opportunity to customize and optimize these models. Furthermore, accurate segmentation of the diseased portion from the leaf background remains a challenge, especially under noisy or ambiguous conditions.
FIGURE 1: Generalized Neural Network Configuration To address these challenges, this paper proposes a comprehensive framework that combines robust segmentation with an optimized deep learning model. The key contributions are: 1. Application of the Fuzzy Rough C-Means (FRCM) clustering algorithm for effective segmentation of diseased regions, leveraging its strength in handling uncertainty and image noise. 2. A novel hybrid classification model where the architecture of a VGG16 CNN is optimized using the Antlion Optimization (ALO) algorithm to enhance its performance for the specific task of cotton disease identification. 3. A comparative evaluation demonstrating that the proposed ALO-VGG16 model achieves superior accuracy compared to several existing methods.
The remainder of this paper is structured as follows: Section 2 details the proposed methodology, Section 3 presents the experimental results and discussion, and Section 4 concludes the work and suggests future directions.
Conclusion
AND FUTURE WORK This research presented an optimized AI-based framework for the automated detection and classification of cotton leaf diseases. The methodology combined robust pre-processing using a Median filter, precise segmentation via the Fuzzy Rough C-Means (FRCM) algorithm, and a hybrid deep learning model. The key innovation was the enhancement of a VGG16 convolutional neural network using the Antlion Optimization (ALO) metaheuristic to optimize its architectural hyperparameter. The proposed ALO-VGG16 model achieved an average classification accuracy of 93.33%, demonstrating superior performance compared to several benchmark models including standard CNN, ResNet, SVM, and ANN. This work validates that integrating nature-inspired optimization algorithms with deep learning architectures can significantly improve the precision of agricultural diagnostic systems.
For future work, the model can be tested on larger, more diverse datasets encompassing multiple crops and diseases captured under real-field conditions with complex backgrounds. Further optimization could explore tuning other hyperparameters (learning rate, filter sizes) using ALO or other metaheuristics. Finally, deploying the trained model as a user-friendly mobile application would be a practical step toward making this technology accessible to farmers for rapid, in-field disease diagnosis. CONFLICT OF INTEREST The authors declare no conflict of interest.
