Dual-frequency identification sonar (DIDSON)can get clear video data from cloudy and dark underwater.To carry out object detection on fish swarm video captured by DIDSON is the precondition of subsequent tracking and recognition.First,we analysed the noise feature of DIDSON image and the statistical property of fish bodies’brightness.Based on that we presented and realised a sonar fish swarm detection method,as well as the suppression method in regard to ghost problem in detection.Finally,we employed the morphological filtering to remove the solitary noise points.Experimental results showed that the proposed method was obviously superior to classic methods in the aspects of sharpness of fish object contour and the accuracy of fish size.It also satisfied the application requirement of real-time tracking and recognition.%双频识别声纳(DIDSON)能在浑浊黑暗的水下获得清晰的视频数据。对双频识别声纳拍摄的鱼群视频进行目标检测是后续跟踪、识别的前提。首先分析双频识别声纳图像噪声特性和鱼体亮度统计特性;基于此,提出并实现一种声纳鱼群检测方法,并就检测中的鬼影问题提出了抑制方法;最后利用形态学滤波去除孤立噪点。实验结果表明,该方法在鱼体目标轮廓清晰度、鱼体大小准确度方面明显优于经典方法,且满足实时跟踪、识别的应用要求。
展开▼