Research On Underwater Image Processing And Object Detection For Seafood

In recent years,as people’s demand for seafood has increased,the aquaculture industry has achieved vigorous development.In order to reduce the intensity of manual labor and improve the intelligent development of marine ranches,it is urgent to carry out research on automated fishing technology for seafood.Object detection is the basis for underwater robots to realize automated detection and capture operations,and has considerable prospects for the intelligent development of marine ranches.The complex underwater environment causes color cast,fogging,and loss of details in underwater images,which reduces the accuracy of object detection.Therefore,this article takes the object detection of seafood as the research background,and conducts research from four aspects:image denoising,image enhancement,object detection and visual detection system design:(1)In view of the fact that a large amount of background noise is easily introduced during the acquisition and transmission of underwater images,and the traditional denoising algorithm has the problem of loss of edge information,this paper proposes an underwater image denoising method based on BF-EPLL.Firstly,to eliminate redundant background noise,the method uses the bilateral filtering BF algorithm to smooth the image;secondly,based on the EPLL theory of the likelihood of the image block,the Gaussian mixture model is used to learn the prior information of the image block;finally,the best matching prior information learned by the model is applied to the output image after BF processing to remove the Gaussian noise introduced in the transmission process.Experiments showed that the denoising algorithm proposed in this paper can effectively overcome the problem of edge information loss in traditional algorithms,and can improve the engineering applicability in the field of underwater image denoising.(2)Aiming at the problems of color shift,fogging and loss of details in underwater images caused by the light absorption and scattering effects of water bodies,this paper proposes an underwater image enhancement method based on image fusion.This method first designs different versions of the input image,and executes the white balance algorithm and the improved dark channel prior algorithm on the original image respectively,aiming to remove the color cast of the image and enhance its contrast;then,design multiple feature weight maps of the two input images,and calculate their normalized weight maps to emphasize the salient features of the target;finally,the multi-scale fusion technology is used to perform feature fusion on the two normalized weight maps to obtain the output map.After experimental testing,the algorithm proposed in this paper can overcome the defects of over-or under-enhancement in traditional fusion algorithms,highlight the salient features of the target,and have higher robustness.(3)In view of the misdetection and missed detection of seafood in complex background in the original YOLOv3 object detection method,this paper proposes an underwater object detection algorithm based on the improved YOLOv3 model.First of all,according to the characteristics of the scale diversity of the target to be detected,the YOLOv3 model is used to realize the multi-scale prediction of the target;second,in order to solve the problem of the target missed detection of the original model,the spatial pyramid network is added to the Darknet-53 network to accquire the fine-grained network structure and improve the detection accuracy;third,in view of the high false detection rate of the original algorithm,the bounding box regression loss function is improved to reduce the error rate of bounding box prediction;finally,the improved network model is trained and tested.Experimental results show that the improved YOLOv3 algorithm can accurately identify targets of different scales in a complex background,reduce the misdetection and missed detection rate of seafood,and has higher generalization and robustness.(4)In view of the low degree of automation in seafood capture,this paper designs an underwater visual object detection system based on machine vision.It mainly completes the selection of various components of the hardware system and the design of the function and detection process of the software system to provide necessary software and hardware support for subsequent robotic automated fishing operations.