针对当前船舶检测模型复杂度高、对设备资源要求较高等问题,提出一种基于YOLOv5s的船舶目标检测算法SRE-YOLOv5s.该算法使用ShuffleNetV2轻量级网络替换YOLOv5s原始特征提取主干网络降低模型复杂度,并使用感受野模块(Receptive Field Block.RFB)增强算法特征处理能力,考虑到Complete IoU(CIoU)损失函数会影响模型收敛速度,采用Efficient IoU(EIoU)损失函数进行优化.通过对SRE-YOLOv5s算法进行训练和验证,结果表明:SRE-YOLOv5s在轻量化的同时保证了检测精度,与其他主流轻量化检测模型相比,SRE-YOLOv5s具有更好的检测性能.此外,根据船舶可视化检测结果可知,SRE-YOLOv5s模型在实际场景中具有较大的应用潜力.
Lightweight Ship Detection Algorithm Based on Improved YOLOv5s
Aiming at the problems of high complexity of the current ship detection model and high re-quirements for equipment resources,a ship target detection algorithm SRE-YOLOv5s based on YOLOV5s was proposed.In this algorithm,ShuffleNetV2 lightweight network was used to replace YOLOv5s original feature extraction backbone network to reduce the model complexity,and RFB was used to enhance the feature processing ability of the algorithm.Considering that the loss function of Complete IoU(CIoU)will affect the convergence speed of the model,the loss function of Efficient IoU(EIoU)was adopted for optimization.By training and verifying the SRE-YOLOv5s algorithm,the results show that SRE-YOLOv5s is lightweight and ensures the detection accuracy.Compared with other mainstream lightweight detection models,SRE-YOLOv5s has better detection performance.In addition,according to the visual inspection results of the ship,the SRE-YOLOv5s model has great ap-plication potential in the actual scene.
ship detectionYOLOv5slightweight networkreceptive field
NETL
NSTL
万方数据
