Research On Deep Learning Object Detection Algorithm Based On YOLO

Computer vision is a long-standing development direction in the branch of computers,but the development of computer vision is slow due to limitations of hardware devices and technologies.With the improvement of computing power and the birth of deep learning,computer vision has become a hot development direction,and object detection is an important development direction in computer vision.Object detection uses computers to automatically locate targets and recognize targets,and it has its own significance in drone tracking and automatic driving.At present,the convolutional neural network in deep learning is the mainstream method of object detection.Due to its inherent characteristics,convolutional neural network has the advantage that traditional object detection is incomparable in image feature extraction.From the originator R-CNN of neural network object detection,to the recent M2 Det,the object detection progresses remarkably,but the existing target detection method cannot balance the accuracy and speed.How to perform object detection well and fast is still an urgent problem,and there is still space for improvement in object detection.Based on the YOLO algorithm,this paper proposes an improved sliding window algorithm to detect objects in high-resolution images by sliding windows of different scales on the input image,improving the accuracy of small object detection,and using image saliency and ResNet identification network.Object position correction and object category correction are performed on the detection results.Finally,the two algorithms are docked to further improve the detection effect of the YOLO algorithm.In this paper,on the remote sensing aircraft image and remote sensing vessel dataset,the improved algorithm proposed in this paper is compared with current mainstream object detection methods,and the accuracy recall curve is used as the evaluation index to verify the effectiveness of the algorithm.