Research And Implement Of Machine Vision Based Underwater Dynamic Fish Size Measurement Method

The growth status of farmed fishes is of great significance for aquaculture.Fishes are put out of water to be measured in traditional manual method,which not only takes time and effort,but also has a negative impact on fish growth.With the continuous development of the aquaculture,how to conveniently and efficiently obtain the growth status of aquatic products has become an urgent problem to be solved in the management of aquatic products.Therefore,this paper focuses on the problem of underwater dynamic fish size measurement based on machine vision in aquaculture environment.The main works of this research are as follows.(1)In underwater environment,the refraction will influence the imaging model when capture the image of fishes,which may cause error in underwater stereo vison application.To solve this issue,an improved underwater calibration method is proposed based on the underwater calibration of single camera combined with flat refractive geometry.The calibrated parameter of underwater binocular is more suitable for underwater stereo vision application,which can achieve higher accuracy.(2)To acquire the fish size of underwater swimming fishes,a fish dimension measurement method based on ellipsoid fitting is proposed,which can calculate the size of multiple fishes from image.The proposed method achieves fish detection and initial segmentation with Mask-RCNN.A segmentation optimization method based on GrabCut is proposed to finetune the result around fish edges.Then,the points cloud of fishes can be extracted according to fish detection and segmentation result.A method based on ellipsoid fitting is proposed to calculate the dimension of fish with various position.It is shown that the proposed method not only has a wide application range,but also has high measurement accuracy,which can meet the needs of aquaculture fish size measurement.(3)An underwater dynamic fish size measurement system is implemented to apply the proposed fish size measurement method on aquaculture environment.A portable underwater binocular camera is developed to capture the video of farmed fishes.The captured video is upload to remote cloud server by Raspberry Pi.The video is analyzed and processed in the cloud server and the sizes of fishes are computed.A GUI application is achieved to provide interfaces for interaction,system maintenance and displaying the processed results.