The Applied Research Of One Stage Object Detection Algorithm Based On Deep Learning

As a popular research direction in the field of machine vision,target detection plays an important role in image content understanding and scene analysis tasks,and is a necessary prerequisite for a large number of advanced visual tasks.Target detection technology is widely used in various digital image processing fields such as traffic road sign recognition,intelligent security system,military target reconnaissance monitoring,medical surgical instrument navigation,etc.It provides reliable assistant decision-making for the computer to understand the current scene content and make the next working state.The target detection algorithms used for different target detection application scenarios are also different,but the core idea is to locate and classify the target,and the target location and classification are achieved by extracting the target features.The completeness and effectiveness of the target feature largely determine the effectiveness of the target detection.At present,the target detection algorithm based on deep learning can effectively solve the problem of target feature completeness,but the effectiveness of the target feature is largely affected.The constraints of the application scenario result in the target detection task in the real scene often failing to achieve the expected effect.Therefore,it is of great theoretical and practical value to study the target detection algorithm for solving practical application problems.With the continuous advancement of observation technology,the image data of various fields such as transportation,military,medical and so on have appeared in the trend of large quantity,high quality and rich information.The data processing results obtained by manual interpretation or traditional target detection technology are not accurate.Sexuality and timeliness,target detection in complex background still has the problem of low precision and slow speed.In this paper,we use the deep learning-based first-order target detection algorithm YOLO(You Only Look Once)to propose a series of lifting detection speeds for the problem of ship detection and high-resolution remote sensing image target detection in real scene.Accuracy,reduce the false alarm false alarm optimization method,and effectively improve the accuracy and practicability of the target detection model.The main research contents of this paper are as follows:(1)Research on first-order target detection algorithm YOLO based on deep learning and its improved method.In this paper,the YOLO algorithm and its improved method are summarized and summarized.As an algorithm known for its detection speed,YOLO has achieved the best target detection effect.This paper introduces and analyzes the algorithm from the perspectives of core ideas,improved methods,model network structure and loss function,and summarizes various optimization methods for improving target detection speed and accuracy.Through careful study of the target detection algorithm,it not only provides powerfultheoretical support for the model improvement method,but also proposes a reasonable solution to the actual problem.(2)In order to solve the problems of low precision,high false alarm rate and poor detection of small target in the target detection task,this paper proposes a second classification algorithm for ship target based on YOLOv2 for the ship target detection scene in aerial data.And use the self-made ship dataset to establish a ship target detection system.The proposed algorithm proposes an improved model for ship detection.By introducing the residual structure,the network deepening model is used to improve the detection ability of small targets.On the basis of the model,various optimization methods such as target secondary classification,model pre-training,and candidate frame dimension clustering are used to improve the performance of the model,realizing the real-time detection of ship targets while reducing false alarm rate and improving target detection accuracy..The experimental results show that the proposed algorithm has higher detection accuracy than the original YOLOv2 algorithm and SSD(Single Shot Multi Box Detector)algorithm.(3)Aiming at the problem of slow and low efficiency of high-resolution remote sensing image target detection,a high-resolution remote sensing image target detection algorithm based on target saliency and improved YOLOv3 is proposed.The algorithm firstly uses the target saliency model to adaptively cut the sample data,filters out the useless background information in the wide-angle large-size high-resolution remote sensing image,and avoids the target of scale normalization by rough extraction of the target potential sub-region.Loss of resolution.After the saliency processing,the algorithm proposes a large-to-medium-sized target in the targeted prediction model based on multi-scale prediction.The pre-training method different from the original algorithm is used to enhance the detection ability of the model for each resolution.On the basis of the existing models,the model is improved by using the candidate box dimension clustering and the optimized network training method.The candidate frame dimension clustering and multi-scale prediction can speed up the model convergence speed and effectively improve the target.positioning accuracy.The proposed algorithm aims to ensure the detection accuracy and speed up the target detection speed by reducing the amount of calculation.The experimental results are compared with other algorithms.The results show that the algorithm can quickly perform target detection under the condition of satisfying high target detection accuracy.,with good target detection capabilities.