Research On Underwater Garbage Detection Based On Image Enhancement And Improved YOLO

At present,ocean lakes and rivers are becoming increasingly polluted,and it is common to see reports that underwater garbage,such as discarded fishing gear,improperly recycled packaging,discarded plastic bags and soda bottles,is threatening biodiversity.Some of the underwater garbage floats on the surface of the water,but most of it sinks to the bottom.Recent studies show that hundreds of thousands of aquatic mammals die every year from pollution caused by underwater garbage,which has a very bad impact on the underwater ecological environment.Therefore,how to remove garbage,protect underwater ecological environment,and maintain the growth of underwater animals and plants has become a common problem faced by people.Due to the particularity of underwater environment,human beings cannot act in the underwater environment as they do on land.It is necessary for divers trained to remove underwater garbage.However,due to poor visibility in underwater environment,it is difficult for divers to see and locate garbage,and some underwater garbage,such as batteries,chemical drums or sharp objects,may pose a threat to the safety of divers.It is difficult to achieve an efficient and sustainable underwater garbage cleanup,which is timeconsuming and labor-intensive.With the development of industrialization,it has become an effective way to clean up underwater garbage by using underwater robot to detect and remove garbage.Underwater target detection is one of the key technologies in the underwater garbage cleaning robot,which has a crucial impact on the efficiency,safety and reliability of the robot.The technology can effectively identify and locate underwater garbage by extracting features from underwater garbage targets,providing necessary information for follow-up tracking or cleaning missions.Traditional underwater target detection methods need to mark the image target manually,which is difficult to adapt to the complex and changeable underwater environment.This paper studies underwater garbage target detection based on deep learning.This paper focuses on investigating underwater garbage,such as discarded cans and plastic bags,as well as biological targets like fish,through the utilization of the UIEB and Trashcan data sets.The research approach involves examining image enhancement and target detection techniques in order to obtain valuable insights.The main areas of focus in this paper include:(1)Due to the absorption and scattering of light by water,there are problems of color deviation,blur and low brightness in underwater images.Consequently,the precision of identifying targets underwater is insufficiently low.In view of the above problems,a fusion imaging model based on deep learning and a fast underwater image enhancement algorithm based on Retinex theory are designed in this paper,named LMIR algorithm.This algorithm uses the depth separable convolutional neural network pseudosynthetic image model and Retinex theory to restore underwater images.Imaging model is used to eliminate scattering and absorption,improve image contrast and correct color bias.Using Retinex theory to improve image illumination;Finally,the two are fused on the channel to obtain the final enhanced image.The experimental results show that the proposed algorithm has better subjective visual effect.It is the highest in PSNR,SSIM and UIQM,and other indicators are also competitive.In terms of speed,FPS is higher than most algorithms,the reasoning speed is faster and the number of parameters is smaller.LMIR algorithm ensures the real-time performance of the whole algorithm and has a good enhancement effect.(2)Nowadays,underwater garbage data sets are relatively scarce and of low quality,and the underwater environment is complicated and the target background is confused.Adapting general target detection algorithms to the underwater environment has been challenging,leading to reduced detection accuracy.Meanwhile,models with high accuracy encounter issues with slow speed due to their large size.To overcome these limitations,this paper introduces the CG-YOLO target detection algorithm,based on the CBAM attention mechanism.The proposed algorithm,also known as Ghost-YOLO,addresses the challenges of underwater target detection and offers improved speed and accuracy.The model is improved on the basis of YOLOv5 s model: Firstly,the underwater image enhancement method proposed in this paper is used to improve the quality of the input image.Secondly,lightweight CBAM attention mechanism module is added to the model detection part to help the model highlight the key information of the target in the complex underwater environment and enhance the model detection ability.Ultimately,the Ghost convolution module is employed to substitute the typical convolution module within the Neck layer of YOLOv5.This replacement effectively reduces the model’s parameter count,resulting in improved detection speed.The experimental results show that the CG-YOLO model can achieve the highest detection accuracy of 93.7% on the image resolution of 640×640,and the calculation time is only 6.1ms,the detection accuracy is high and the real-time performance is satisfied.(3)With the application of deep learning in underwater,relevant researches have developed rapidly.However,most of them only stay at the academic level and are difficult to be put into practice.In order to realize the usability of the improved algorithm,this paper deploys CG-YOLO on Nvidia Jeston Nano,an embedded device with limited resources,uses the underwater camera to shoot the simulation data set for model training,and uses the CG-YOLO algorithm on Nvidia Jeston Nano to realize the real-time detection of underwater garbage.The model deployment is successfully implemented and the usability of the model is verified.The research results of this paper have practical significance for underwater environment protection.A fast and effective solution for underwater garbage detection is proposed,which provides a theoretical basis for underwater garbage cleaning.Finally,the deployment of the edge equipment and the use of the algorithm are realized,which has a good reference for the research and development of underwater garbage cleaning robot.