Mobile Object Detection Algorithm based on YOLOv5

• 1、Electronic Engineering School,Beijing University of Posts and Telecommunications,Beijing 100876

Abstract：With the popularization of mobile devices, the application field of mobile phones is becoming increasingly diverse.However, the use of mobile phones is restricted in some specific places. Due to small object and high background interference, mobile phones are prone to false alarms and missed detections during the detection process. To address these challenges, this paper proposes improvements to the YOLOv5s algorithm for mobile phone object detection.The algorithm designs a focal loss-based loss function friendly to minority samples, which adjusts the weights of samples to address the problem of a relatively large proportion of mobile phone objects and a small number of similar interfering objects in the dataset, thereby improving detection accuracy. The traditional NMS method is also improved by enhancing the confidence of mobile phone bounding boxes, uniformly comparing bounding boxes of different categories, and reducing the IoU threshold for determining overlapping areas of mobile phone objects. This effectively increases the weight of mobile phone objects and reduces missed detections.Experimental results demonstrate that the proposed improvements achieve satisfactory results on self-collected mobile phone detection data, providing strong support for mobile phone detection.

Keywords： Computer Applocation Technology Mobile Phone Object Detection Minority-friendly Loss Function Non-Maximum Suppression