Optimizing Mango Leaf Disease Classification Using VGG16 with Grouped Convolution for Improved Accuracy and Efficiency

The application of VGG16 in detecting diseases in mango leaves is a key area of interest for improving agricultural productivity. This study aimed to evaluate the performance of VGG16 integrated with grouped convolution for detecting mango leaf diseases. Specifically, the research sought to assess the impact of varying the number of groups in grouped convolution on classification accuracy and computational efficiency. A quantitative approach was used, where VGG16 was modified with grouped convolution to classify mango leaf diseases into multiple categories. Various configurations of grouped convolution (2, 4, and 8 groups) were tested on a dataset of mango leaf images. The performance of these models was compared to previous studies using traditional CNN models. The model incorporating VGG16 with 8-grouped convolution achieved the highest accuracy of 98.67% for classifying three disease categories. It outperformed other models, such as XGBoost and GoogleNet, in both classification accuracy and stability, particularly in multi-class settings. Grouped convolution enhanced model efficiency by reducing the number of parameters without sacrificing performance, achieving better results than other traditional CNN architectures such as AlexNet and CNN. The integration of grouped convolution into VGG16 significantly improved classification accuracy and computational efficiency for mango leaf disease detection. The results confirm that grouped convolution is an effective modification for handling multi-class image classification in agriculture. Future studies should explore dynamic grouping strategies and hybrid attention mechanisms to further enhance the model's adaptability and performance in diverse conditions. This study aims to detect early using VGG16 with grouped convolution, enabling farmers to take quick action and increase the productivity and quality of agricultural products.