针对在皮肤黑色素瘤目标检测中出现精度低、漏检率高等问题,提出了一种基于改进YOLOv7 的皮肤良恶性病变检测算法.首先,在头部网络中,采用GSConv卷积结构代替网络模型中的普通卷积结构,降低模型权重,提升对皮肤良恶性病变的检测精度;其次,在骨干网络中融合SE(squeeze and excitation)注意力机制,以提升对皮肤病变的特征提取能力;最后,利用EIOU损失函数代替CIOU损失函数,进一步提高检测精度.实验表明,经过改进后的YOLOv7模型在皮肤良恶性病变数据集上的平均检测精度均值@0.5(mean average precision@0.5,mAP@0.5)达到了90.9%、精确度为89.1%,较YOLOv7 模型平均检测精度均值提升了4.4%、精确度提升了11.0%、运行参数较YOLOv7 降低了11.6%.所提出的改进的YOLOv7 模型能够很好地识别良恶性病变,能够更好地辅助医生诊断.
Detection Algorithm of Benign and Malignant Skin Lesions Based on Improved YOLOv7
In response to the problems of low accuracy and high missed detection rate in target detection of skin melanoma,this paper proposes a skin benign and malignant lesions detection algorithm based on improved YOLOv7.Firstly,in the head network,the GSConv convolution structure is employed as a substitute for the con-ventional convolution structure in the network model.This substitution reduces the model's weight while enhan-cing the detection accuracy of benign and malignant skin lesions.Secondly,the SE(squeeze and excitation)at-tention mechanism is fused in the backbone network to enhance the feature extraction ability of skin lesions.Lastly,the EIOU loss function is utilized in place of the CIOU loss function to further enhance the accuracy of detection.Experimental results demonstrate that the enhanced YOLOv7 model achieves an average detection pre-cision mean(mAP@0.5)of 90.9%and an accuracy of 89.1%on the dataset of skin benign and malignant le-sions.The proposed improved YOLOv7 model demonstrates a significant advancement in performance,with a 4.4%increase in average detection precision mean and 11.0%improvement in accuracy compared to the baseline YOLOv7 model.Additionally,the operating parameters have been reduced by 11.6%in comparison to YOLOv7.This enhanced model effectively identifies benign and malignant lesions,offering valuable assistance to doctors in the diagnosis process.
NETL
NSTL
万方数据
