Lska-Yolo: Improved Yolo Framework Tailored for Cervical Cell Detection
Jiayi Chen, Le Chen, Wenming Yu, Yuanfang Qiu, Hongmian Li · 2025
Cervical cell detection is of great significance in the early screening of cervical cancer, but traditional detection methods have problems such as low efficiency and susceptibility to subjective factors. In recent years, target detection techniques based on deep learning have provided new ideas to solve this problem. However, due to the diverse cell morphology and complex background in cervical cell images, the existing detection models still face the challenge of insufficient accuracy. To address this problem, this paper proposes an improved cervical cell detection model LSKA-YOLO based on YOLOv8 and Large Separable Kernel Attention (LSKA) mechanism. The model combines the wide convolutional kernel with the large receptive field and the high efficiency of separable convolution, it can effectively enhance the feature extraction ability of the model for cell targets and reduce the computational complexity at the same time. Validated by a large number of comparative experiments, LSKA-YOLO performs well in the cervical cell detection task, with Recall reaching 81.9 % and mAP improving to 85.1 %, which are$\mathbf{5. 1 \%}$and$\mathbf{1. 6 \%}$higher compared to YOLOv8, respectively. The experimental results show that the LSKA-YOLO model, which is able to capture the key features in cervical cell images more effectively while suppressing the background noise and interfering information, significantly reduces the phenomena of false detection and missed detection. The model performs well in the cervical cell detection task, with substantially improved detection accuracy and robustness, providing efficient and accurate technical support for early screening and diagnosis of cervical cancer.