一种目标检测模型训练方法、装置、设备及存储介质
By screening and constructing an effective negative sample set, the target detection model is trained, which solves the problems of false detection and overfitting in existing technologies and improves the model's generalization ability and detection accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JINAN BOGUAN INTELLIGENT TECH CO LTD
- Filing Date
- 2022-10-09
- Publication Date
- 2026-07-17
AI Technical Summary
Existing object detection models suffer from overfitting and insufficient generalization ability, especially in pedestrian detection, where they are prone to misdetecting non-target objects such as dogs. Existing methods fail to effectively utilize negative samples for training, resulting in overfitting of the model on repetitive scenes or targets and poor training performance.
By acquiring initial negative samples, the target detection model to be optimized is used for detection, and an effective negative sample set is selected. Based on this set, a target training set is constructed, and the target detection model to be trained is trained. Mosaic data augmentation technology and image replacement operation are used to enrich the diversity of training samples.
It improved the model training effect, reduced the false detection rate, enhanced the model's detection capability in different scenarios and targets, and prevented the model from overfitting in repetitive scenarios.
Smart Images
Figure CN115546693B_ABST