一种海上养殖网箱的无人巡检方法研究

针对海上养殖网箱人工巡检的成本高、风险大的问题,以海上河鲀养殖网箱监测为应用背景,提出了一种无人化网箱巡检的新方法,该方法利用无人机俯视观测法,并结合日常网箱养殖状态信息,可高效地完成基础养殖数据采集任务。在无人机拍摄的视频数据基础上,提取关键帧,利用卷积神经网络HED(Holistically-nested edge detection )进行边缘检测作为图像预处理,起到降低图像冗余信息,得到清晰网箱边缘信息的作用,在此基础上将边缘图二值化并提取目标区域,提出了相应的自适应阈值选取规则,最后根据改进的Tamura纹理特征对网箱养殖区进行数据有效性的判断。该方法结合深度学习方法与传统图像检测技术,具有较强的环境自适应性和较高的准确性。最终以大连天正实业有限公司大李家红鳍东方鲀养殖场的养殖网箱作为实验对象,海上河豚养殖网箱提取的准确度为97%,信息的有效性判断准确度为97.1。

The Study on a Method of Unmanned Inspection for Marine Aquaculture Cage

Aiming at the problem of high cost and high risk of manual inspection of cages in marine aquaculture, a new method of unmanned cage inspection is proposed based on the application background of monitoring of cages in marine aquaculture. This method can efficiently complete the data collection of basic aquaculture by using UAV overlooking observation method and combining with the information of daily cage culture conditions. On the basis of video data captured by UAV, key frames are extracted, and image preprocessing is carried out by using the Holistically-nested edge detection network (HED), which reduces the redundant information of the image and obtains the edge of the cage. On this basis, the edge image is binarized and extracted the target region , and the corresponding adaptive threshold selection rules are proposed. At last, according to the improved Tamura texture features, the validity of the data in the cage farming area is judged. This method combines depth learning method with traditional image detection technology, and has strong environmental adaptability and high accuracy. Finally, the aquaculture cage of Dali Red Fin Oriental Fugu Farms in Dalian Tianzheng Industrial Co., Ltd. was taken as the experimental object. The accuracy of extracting the aquaculture cage of puffer dolphin at sea was 97%, and the accuracy of judging the validity of the information was 97.1%.