Research on Target Detection of Pine Flower Images Based on Deep Learning

This study aims to achieve automated detection of pine flowers using deep learning technology, addressing the current issues of high cost, low efficiency, and safety risks associated with manual collection of pine pollen. By combining field collection, online searches, and data augmentation techniques, a dataset containing 1,344 pine flower images was constructed, and the images were manually annotated using the LabelImg tool. Based on this, three versions of the YOLO series target detection algorithms—YOLOv5, YOLOv8, and YOLOv11—were selected to train corresponding models, verify their detection performance, and comprehensively analyze their applicability differences on this dataset. Experimental results show that YOLOv11m performs best in terms of precision, achieving a score of 1, while YOLOv5m excels in recall and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{mAP}^{50-95}$</tex> metrics, reaching 1 and 0.959, respectively. Overall, both YOLOv5m and YOLOv11m achieve a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{mAP}^{50}$</tex> score of 0.995, demonstrating superior detection performance. Although YOLOv5m has a slight advantage in some metrics, the three models all exhibit high levels of precision, recall, and mean average precision (mAP), with no significant overall performance differences. The results indicate that YOLOv11m, while meeting the resource constraints of embedded systems, can ensure high detection performance, making it suitable for deployment in resource-limited devices and providing technical support for the automation of the pine flower industry.