ENHANCED MASK REGIONAL BASED CONVOLUTIONAL NEURAL NETWORK USED FOR DETECTING PERCENTAGE OF MAIZE LEAF DAMAGE

Abstract

In this work, an Enhanced Mask R-CNN-based segmentation approach combined with an Adaptive Median Neural Network (AMNN) for denoising is provided for accurate detection and quantification of pest-affected maize leaf regions. Pest infestation is one of the major challenges in maize crop production, resulting in great yield losses. Four major pests, namely armyworms, aphids, grasshoppers, and beetles, are analyzed on datasets from Roboflow and Mendeley. This Paper using 4 datasets for finding the maize leaf affected percentage. From the segmentation results, shows that the highest percentage of leaf damage was caused by armyworm followed by aphids, grasshoppers, and beetles. The exact quantification of damage was done by calculating the area affected based on pixel-wise segmentation metrics. In the proposed method, high segmentation accuracy reduces the false positives and improves the identification of pests. Performance evaluation using accuracy, precision, recall and f-measure shows model reliability. In these findings the proposed segmentation method achieves 99.53 accuracy values compared to other existing methods. Future enhancements include real-time deployment with IoT-based monitoring systems and hyperspectral imaging for further improving the accuracy of detection. It will contribute to precision agriculture, helping farmers in early pest detection and efficient pest management strategies.