Enhanced YOLO12 with spatial pyramid pooling for real-time cotton insect detection

TL;DR

This paper introduces Enhanced YOLO12, a deep learning model for real-time detection of cotton insects, offering improved accuracy and efficiency for sustainable pest management.

Contribution

The paper introduces Enhanced YOLO12, a novel deep learning model with optimized spatial pyramid pooling and attention-based features for improved insect detection.

Findings

Enhanced Hybrid YOLO12 achieved 0.942 precision and 0.876 recall for cotton insect detection.

The model outperformed standard YOLO12 with higher mAP50 (0.945 vs. 0.913) and mAP50-95 (0.735 vs. 0.662).

The model is suitable for real-time pest management in precision agriculture.

Abstract

Effective insect detection is crucial for sustainable cotton production, yet traditional monitoring methods remain labor-intensive, inefficient, and environmentally detrimental. This study introduces Enhanced YOLO12, a novel deep learning architecture for real-time cotton insect detection. Building on the YOLO12 framework, the proposed model integrates an optimized Spatial Pyramid Pooling (SPP) module and attention-based feature extraction to improve detection accuracy while maintaining computational efficiency. To ensure robustness, we developed and evaluated multiple baseline models (standard YOLO11 and YOLO12) and custom architectures (YOLO12_Fusion, YOLO11-BRA-Net, YOLO11_CBAM, and Enhanced Hybrid YOLO12). According to the conducted experiments, Enhanced Hybrid YOLO12 achieved the best performance, achieving 0.942, 0.876, 0.945, and 0.735 in precision, recall, mAP50 and mAP50-95, respectively. It significantly outstands the results of the standard YOLO12 (0.925, 0.848, of 0.913, and 0.662). These results demonstrate that Enhanced Hybrid YOLO12 can be considered as a state-of-the-art framework for precision agriculture, with its high detection accuracy and real-time capability. Therefore, they encourage this deep learning model in pest management applications.

Figures