Research On The Method Of Image Processing And Object Detection In Amphibious Environment

The amphibious environment includes underwater,land and water-land junction areas.The environment of the underwater and water-land junction areas is complicated.When the robot operates in the amphibious environment,there are often problems such as signal loss and positioning failure;in this case,the robot will be offline State.If the robot cannot perceive its environment,surrounding targets,scenes,etc.when offline,it will greatly affect the robot’s operational capabilities and put the robot in a dangerous state: such as missed and misrecognized targets,Causes the robot’s tracking task to be interrupted;unable to perceive its own position and motion status,resulting in the inability to perform accurate motion control on the robot;unable to effectively observe the work scene,resulting in the robot being unable to make decisions and judgments based on the current robot status,so the robot must be improved Perception ability in an amphibious environment.However,the image processing problems in the amphibious environment are too complicated due to the image color degradation,lack of illumination,water surface occlusion,and scattering of underwater particles in the area between underwater and land and water,and it has not been well solved so far.It makes the robot’s perception in the amphibious environment full of challenges.In this paper,aiming at these problems,through an image processing algorithm and target detection algorithm suitable for amphibious environment,research methods to improve the environment perception ability of robots in amphibious environment.The main research contents of this article are:End-to-end amphibious image classification algorithm.Take the statistical characteristics of the image data as the basis: take the statistical characteristics of the image data,contrast,sharpness,brightness,mean,and variance as input,and fit them with the image category through a neural network,that is,underwater image,land image or water The mapping of the part of the image in the empty junction realizes a semi-supervised learning algorithm for improving pseudo-labels through data similarity,and builds a semi-supervised amphibious environment image classification algorithm based on the improvement of data similarity.Carry out comparison experiments with pseudo-label algorithms,data similarity experiments,comparison experiments with image classification algorithms based on VGG16,and data labeling ratio experiments to verify the effectiveness of the algorithm in this paper and solve the problem of scene recognition in amphibious environments.Amphibious environment image preprocessing algorithm based on Actor-Critic deep reinforcement learning algorithm.Through the use of the amphibious environment image preprocessing toolbox by the agent,which includes tools for image denoising,image color preprocessing,and image blurring,an amphibious environment image processing system is realized.Carry out comparison experiments with dark channel prior method,histogram equalization method,contrast-limited adaptive histogram equalization method,multi-scale retinal cortex algorithm,and unsupervised generation network monocular real-time color correction algorithm to verify the algorithm of this paper Effective,solve the problem of difficult image preprocessing in amphibious environment.By studying the image segmentation method of the unique water-air junction area image in the amphibious environment,the images in the amphibious environment are divided into three categories into two categories;the target data set in the amphibious environment is constructed;the YOLOv4-tiny for target detection in the amphibious environment is studied The algorithm improves the spatial attention mechanism of the original algorithm,so that the neural network can better extract the characteristics of the target in the water-air junction area.A comparative test is carried out to verify the effectiveness of the algorithm in this paper,and solve the problem of difficulty in detecting water surface targets in an amphibious environment.This paper divides the environmental perception problem in the amphibious environment into three parts: scene recognition,image preprocessing and target detection.The scene recognition problem is solved by a semi-supervised learning algorithm based on data similarity,and the image preprocessing problem is solved by a reinforcement learning-based The amphibious environment image preprocessing algorithm is used to solve the problem,and the target detection problem is solved by the improved YOLOv4-tiny algorithm.Through the processing of the environment perception system in this paper,the robot can better perceive the environment and targets around itself in the amphibious environment,and can improve the robot’s operating ability in the amphibious environment.