研究目的
Automatically recognizing and localizing prohibited items in airport X-ray security images to improve the efficiency and reduce security risks of security inspections.
研究成果
The Attention Mechanism based CNN model achieves recognition and location of prohibited items in airport security X-ray images with only weak supervision training. The model can tell the security inspectors the category of prohibited item and remind them where the dangerous goods are, facilitating the reinspection. The model provides a new direction for automated security.
研究不足
The dataset used for training and testing is small and has a monotonous background, which may limit the model's generalization ability. The model requires weak supervision training, but the accuracy of localization could be improved.
1：Experimental Design and Method Selection:
A top-down attention mechanism is applied to enhance a CNN classifier to locate prohibited items. A high-level semantic feedback loop is introduced to map the target's semantic signal to the input X-ray image space for generating task-specific attention maps.
2：Sample Selection and Data Sources:
The GDX-ray dataset is used, which captures single-target and multi-target X-ray security images of guns, knives, darts, and other dangerous goods from multiple angles. Data augmentation is performed to expand the dataset.
3：List of Experimental Equipment and Materials:
Google net pretrained with ImageNet 2012 training set is used as the CNN model.
4：Experimental Procedures and Operational Workflow:
The model is trained using transfer learning strategy. The learning rate of the bottom network is set to
5：0001, and the top network is set to After 30 iterations with batchsize set of 8, the network tends to convergence. Data Analysis Methods:
The model's recognition and location capabilities are verified using the security inspection strategy. The classification accuracy and localization IoU are calculated to evaluate the performance.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
