TANG Yulin, BIAN Shaofeng, ZHAI Guojun, LIU Min, ZHANG Weidong. Improved YOLOv5 Method for Detecting Shipwreck Target with Side-scan Sonar[J]. Geomatics and Information Science of Wuhan University. DOI: 10.13203/j.whugis20210353
Citation: TANG Yulin, BIAN Shaofeng, ZHAI Guojun, LIU Min, ZHANG Weidong. Improved YOLOv5 Method for Detecting Shipwreck Target with Side-scan Sonar[J]. Geomatics and Information Science of Wuhan University. DOI: 10.13203/j.whugis20210353

Improved YOLOv5 Method for Detecting Shipwreck Target with Side-scan Sonar

  • In view of the problems of high miss-alarm rate of small targets, heavy model weight, and detection speed that fails to meet real-time requirements in side-scan sonar shipwreck detection method based on the YOLOv3 model. The paper introduces the YOLOv5 algorithm and proposes a model based on YOLOv5 according to the characteristics of the side-scan sonar shipwreck dateset.Try YOLOv5a, YOLOv5b, YOLOv5c, YOLOv5d, YOLOv5s, YOLOv5m, YOLOv5l and YOLOv5x under the basic framework of YOLOv5 with eight different depth and width model structures.Choose the best structure use GA+K (Genetic Algorithm+K-mean) algorithm to optimize the detection frame, and improve the loss function through CIOU_Loss. The experimental results show that the improved YOLOv5a model is 0.3% and 0.6% higher than the original model in AP_0.5 and AP_0.5-0.9, and has a substantial improvement compared with the YOLOv3 model, in which AP_0.5 and AP_0.5-0.9 are improved by 4.2% and 6.1% and the the detection speed reaches 426 frames per second, which is almost doubled that of YOLOv3, which is more conducive to practical applications and engineering deployment.
  • loading

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return