Research On Image Processing And Target Detection For Underwater Garbag

With the development of society,the problem of underwater trash has affected human life.Underwater trash is widely distributed and manual cleaning is limited and inefficient.Therefore,it is necessary to develop underwater robot to replace manual cleaning of underwater trash.Object detection is the primary problem in underwater environment perception,and a prerequisite for the realization of automatic trash recovery by underwater robots.In this paper,the underwater trash object detection system based on the YOLOv5 model of deep learning is studied.The main research of this paper is as follows.(1)In order to solve the problem of sparse datasets,the trash＿ICRA19 dataset is enhanced in this paper.The data is augmented from 7,667 in the original dataset to 13,665,providing data to support the object detection of the convolutional neural network.An image enhancement algorithm based on underwater imaging model is proposed to solve the problems of insufficient light,image fogging,and low contrast in underwater environmental imaging.Based on the underwater imaging model and dark channel prior algorithm,a method for estimating atmospheric light using only the blue-green channel to calculate the underwater dark channel information is proposed,which enhances the robustness of the model.Simultaneously,an enhancement function is used to adjust the brightness of the enhanced image by the dehazing model while maintaining the natural brightness of the input image.By this way,the problem of uneven underwater brightness is solved and the quality of underwater images are effectively improved.(2)In order to further improve the detection accuracy of underwater trash,an improved YOLOv5 model object detection algorithm is proposed.Firstly,a priori frame of the model is redesigned to achieve a more accurate fit between the priori frame and the true frame.Then,the backbone network of the original model is improved using the ShuffleNetV2 model and MCA attention mechanism module.Finally,to improve the efficiency of position loss convergence,the loss function of the model is improved,which effectively increases the convergence speed of the bounding box and reduces the real-time detection accuracy of small sizes,and reducing the underwater rubbish leakage and false detection rates.The improved model achieved a mean average precision of 92.10% on the trash＿ICRA19 dataset,demonstrating good robustness and generalization,and meeting the requirements of practical applications.(3)In order to build an underwater trash object detection system,a suitable hardware module is selected and a reasonable embedded system software is designed.In this paper,the NVIDIA XAVIER NX was chosen as the embedded control module,and the image preprocessing and target detection algorithm were ported to the embedded control module.The designed underwater trash object detection system was tested in environments with sufficient and insufficient lighting,and compared with other target detection models.The results showed that the system exhibites good real-time detection performance and achieved good detection results.