Global agricultural productivity is seriously threatened by pest outbreaks, underscoring the urgent need for intelligent, effective, and interpretable automated detection systems. This study presents a novel hybrid deep learning framework that integrates YOLOv11 for real-time pest localization and VGG16 for enhanced classification accuracy, with Grad-CAM++ providing visual interpretability to increase transparency and user trust. The framework is trained and evaluated on the IP102 dataset, which includes 102 pest categories and poses challenges such as small object detection, class imbalance, and high intra-class visual similarity. The proposed model achieves an Intersection over Union of 90.8% and a mean Average Precision of 98.3%, with precision, recall, and F1-score values of 97.8%, 98.5%, and 98.2%, respectively. Comparative analysis against Faster R-CNN, SSD, EfficientDet, and YOLOv3 confirms its superior accuracy and inference efficiency. Moreover, the framework delivers real-time performance, achieving 142 FPS on an RTX 3090 and 35 FPS on a Jetson Nano, ensuring feasibility on both high-end and edge devices. By combining accurate detection, interpretability, and high-speed processing, this work makes a novel and practical contribution to precision agriculture.