An Improved Network for Detection and Counting of Cabbage Using UAV Image

Highlights An improved lightweight model for cabbage maturity detection based on deep learning is proposed. A cabbage maturity detection dataset (CMDD) is established for experiments. The implementation of image slicing and counting in large scenes demonstrates the practicality of the algorithm. The performance of the proposed algorithm is effective and practical for detecting small-to-medium target plants in natural scenes. Abstract. Cabbage is a significant economic crop and an essential staple in people’s daily lives, with its closed state serving as a crucial indicator of maturity. Accurately assessing the maturity and quantity of cabbage to determine the optimal harvest time is vital for production. Due to the small-to-medium size of cabbage targets from the drone's perspective and the high similarity in maturity features, we have enhanced the YOLOv8n network by incorporating methods such as the Space-to-Depth Convolution (SPD-Conv), the Deformable Large Kernel Attention (DLKA), and the Dynamic Head (Dyhead). This has led us to propose a novel deep learning network called SDD-YOLOv8 that enhances capabilities in extracting features and performance in detection. The experimental results indicate that our enhanced model attains a mean average precision ([email protected]) of 87.0%, which is a 1.2% improvement compared to the baseline model. Furthermore, we have devised a technique for detecting and counting cabbages in large scenes to enhance practicality. Our method achieves an overall average counting accuracy of 95.0% compared to manual counting, with individual category accuracies reaching 93.9% and 95.0%. This method enables accurate classification and counting of cabbages in natural environments. Keywords: Attention mechanism, Deep learning, Maturity detection, Plant counting, UAV, YOLOv8.