Object Detection Model Based on Attention Mechanism

Detection of objects is a key research direction in computational vision. The purpose is to identify whether an image contains a specific target by employing corresponding detection methods and returning the position of the detected target in the image. This field finds extensive applications in autonomous driving, medical diagnosis, satellite images, and more. Aiming at the problems of the existing models, such as insufficient receptive field and weak detection ability of targets, more noise interference in the feature extraction process, and false detection and missed detection caused by environmental interference, this paper proposes a object detection model based on attention mechanism, and constructs a target detection system based on this. Firstly, this paper introduces the attention mechanism and related data sets. After analyzing the basic framework of You only look Once vision 7 (YOLOv7), it is proposed that attention mechanism should be taken into account after three characteristic graphs are output to its backbone network, and the model structure with the highest accuracy can be obtained through comparative experiments of control variables. The experimental results show that adding attention mechanism can improve the detection accuracy of YOLOv7 network architecture, and the average accuracy of three different models on PASCAL VOC data set is increased by 1.67%, 2.09% and 2.11% respectively.