Lightweight and Efficient Real-Time Tomato Detection Based on Improved YOLOv11 Network

Faced with the challenge of detecting tomatoes in unstructured environments with dense foliage, overlapping fruits, and varying lighting conditions, this study proposes a high-efficiency, lightweight detection algorithm based on YOLOv11 for real-time detection on low-computation platforms. By adopting MobileNetv4 (small version) as the backbone, the model significantly reduces computational costs while maintaining effective feature extraction. Additionally, the Weighted Bi-directional Feature Pyramid Network (BiFPN) is introduced as the neck to enhance feature fusion, improving detection performance in occlusion, and overlapping scenarios. Experimental results demonstrate that the improved model achieves 96.4% mean Average Precision (mAP), 5.3G FLOPs, 2.1M parameters, and an inference time of 6.5ms, striking a balance between accuracy and efficiency. This research lays the foundation for automated tomato detection, contributing to applications in yield estimation, precision agriculture, and intelligent harvesting systems.