An Improved Adaptive Multi-Scale Attention Dung Beetle Convolution Neural Network-Based Apple Leaves Disease Detection

The identification of diseases in apple leaves is vital in modern agriculture. This study presents a method using Convolutional Neural Networks (CNNs) to improve disease detection, addressing errors and fitting issues. The proposed solution, the improved adaptive Multi-scale Attention Dung Beetle CNN (MusD CNN), shows significant advancements. Various datasets, like the apple disease dataset and PlantVillage dataset, provide diverse apple leaf images representing different diseases. Precise segmentation of affected areas on leaves is done using an Effectual Multi-encoder Decoder Multi-scale Convolutional U-Net (EMED-DUNet) model, enhancing detection accuracy. The MusD CNN, employing an Adaptive Multi-scale Residual Attention CNN for classification with optimized activation functions, improves gradient flow and stability. Performance evaluation using multiple metrics demonstrates high accuracy (99.9%), recall (99.9%), precision (99.9%) and F1-score (99.9%). This research offers an automated solution for apple leaf disease detection, supporting precision agriculture and crop health preservation.