Ding Gege, Shi Yuhang, Liu Zhenquan, Wang Yanjuan, Yao Zhixuan, Zhou Dan, Zhu Xuexiu, Li Yiqin
China Waterborne Transport Research Institute, Beijing 100088, China.
School of Railway Intelligent Engineering, Dalian Jiaotong University, Dalian 116028, China.
Biomimetics (Basel). 2025 Jan 16;10(1):62. doi: 10.3390/biomimetics10010062.
The identification and detection of microalgae are essential for the development and utilization of microalgae resources. Traditional methods for microalgae identification and detection have many limitations. Herein, a Feature-Enhanced YOLOv7 (FE-YOLO) model for microalgae cell identification and detection is proposed. Firstly, the feature extraction capability was enhanced by integrating the CAGS (Coordinate Attention Group Shuffle Convolution) attention module into the Neck section. Secondly, the SIoU (SCYLLA-IoU) algorithm was employed to replace the CIoU (Complete IoU) loss function in the original model, addressing the issues of unstable convergence. Finally, we captured and constructed a microalgae dataset containing 6300 images of seven species of microalgae, addressing the issue of a lack of microalgae cell datasets. Compared to the YOLOv7 model, the proposed method shows greatly improved average Precision, Recall, mAP@50, and mAP@95; our proposed algorithm achieved increases of 9.6%, 1.9%, 9.7%, and 6.9%, respectively. In addition, the average detection time of a single image was 0.0455 s, marking a 9.2% improvement.
微藻的识别与检测对于微藻资源的开发利用至关重要。传统的微藻识别与检测方法存在诸多局限性。在此，提出了一种用于微藻细胞识别与检测的特征增强YOLOv7（FE-YOLO）模型。首先，通过将坐标注意力组混洗卷积（CAGS）注意力模块集成到颈部来增强特征提取能力。其次，采用SIoU（SCYLLA-IoU）算法替代原模型中的CIoU（完整交并比）损失函数，解决收敛不稳定的问题。最后，采集并构建了一个包含7种微藻6300张图像的微藻数据集，解决了微藻细胞数据集缺乏的问题。与YOLOv7模型相比，所提方法的平均精度、召回率、mAP@50和mAP@95均有显著提高；所提算法分别提高了9.6%、1.9%、9.7%和6.9%。此外，单张图像的平均检测时间为0.0455秒，提高了9.2%。
