Improving Accuracy of Small Object Detection in Screen based on YOLOv7

Small object detection encounters various challenges, including low resolution, low contrast, missing visual features, object loss and missed detections, computational efficiency, and data imbalance. To address these challenges, a method is provided to enhance the accuracy of small object detection. This method utilizes a model with a small object detection layer to predict the segmented images. Adding small object detection layers enhances the perception of small objects and improves the network's performance in localizing and recognizing small objects. Using image segmentation strategies involves dividing the image into multiple sub-images for object detection, effectively handling small objects. This paper also presents the implementation principles of the YOLOv7 algorithm and the specific definition of small objects, validated through experiments to demonstrate the feasibility of the proposed methods. Finally, the article summarizes the experimental results.