基于计算机视觉的海滩垃圾识别模型的研究

Research on the Beach Litter Recognition Model Based on Computer Vision

我国是一个海洋大国,海洋面积大、分布广、海岸线长。由于人们对海洋环境认识不足,导致海洋垃圾广布于海洋和海滩中,危及生态环境、经济发展和人类健康。海洋垃圾主要来源陆源和海源两部分,而海滩处于陆海交界地带,受流域、海域潮波作用及人类活动的影响,积累了大量固体废弃物。针对上述问题,提出基于计算机视觉对海滩垃圾进行高效、准确、无损目标检测的设想,实现垃圾在线精准目标检测。首先,基于YOLOv5框架来获得海滩垃圾目标检测模型;其次,利用TensorRT加速框架,对所得模型进行算法执行优化,加速模型推理过程,为模型在有限硬件及算力资源的边缘设备平台部署提供模型基础;最后,基于Jetson Nano边缘设备进行模型本地化部署,实现海滩垃圾的在线精准目标检测。

China is a large marine country,with a large ocean area and wide distribution and long coastlines.Due to the insufficient understanding of the marine environment,marine litter is widely distributed in oceans and on beaches,endangering the ecological environment,economic development and human health.Marine litter mainly comes from land sources and sea sources,and the beach is located at the junction of land and seas and has accumulated a large amount of solid waste under the influence of the tidal wave action of river basins and sea areas and human activities.In view of the above problems,this paper puts forward the research on the efficient,accurate and non-destructive target detection of beach litter based on computer vision,completes the process of model reasoning and the deployment of model practice,and realizes the online accurate target detection of litter.This paper first obtains the beach litter target detection model based on the YOLOv5 framework.Then it uses the TensorRT acceleration framework to optimize the algorithm execution of the obtained model and accelerate the process of model reasoning,so as to provide a model basis for the deployment in the edge device platforms of limited hardware and computing resources of the model.Finally,it carries out the localization deployment of the model based on the Jetson Nano edge device to realize the online accurate target detection of beach litter.The research results can effectively reduce counting errors,improve production efficiency and reduce labor costs.