现代电子技术

针对现有X光图像违禁品检测精度低、速度慢且公共数据集样本过少等问题，提出一种基于CycleGAN和改进YOLOv8n的X光图像违禁品检测算法。该算法利用CycleGAN模型生成更多具有安检X光风格特点的违禁品图像，建立足够多的违禁品图像样本；同时，针对X 光违禁品图像目标大小不同的特点，设计了混合卷积注意力（HCA）机制，有效加强YOLOv8n 对目标关键特征的提取能力。对于安检X 光违禁品图像特征信息不足的问题，利用CARAFE 模块替代原YOLOv8n的上采样方式，并采用BiFPN对颈部网络进行改进，通过有效的特征融合方法和上采样方式弥补图像特征信息不足的问题。最后针对YOLOv8n检测头参数冗余问题，使用参数共享和DBB模块减少模型参数量，同时保证模型的检测精度。在经过扩充的SIXary数据集上对所提出的检测算法进行测试，实验结果表明，该算法的mAP值达到93.1%，相比原始算法提升2.9%，同时参数量减少8.5%，并且保持了与原始模型相当的检测速度。该算法的性能相比原始算法取得显著提升，验证了其改进的有效性。

In allusion to the problems of low accuracy, slow detection speed and too few samples in public data set, an X⁃ray image contraband detection algorithm based on CycleGAN and improved YOLOv8n is proposed. In this model, the CycleGAN model is used to generate more contraband images with security X⁃ray style characteristics, and establish enough samples of contraband image. According to the characteristics of different sizes of objects in X ⁃ ray contraband images, the hybrid convolutional attention (HCA) mechanism is designed to effectively enhance the ability of YOLOv8n to extract key features of objects. In allusion to the problem of insufficient feature information of security X⁃ray contraband images, CARAFE module is used to replace the original YOLOv8n upsampling mode, BiFPN network is used to improve the neck network, and the effective feature fusion method and upsampling method are used to made up the insufficient image feature information. Aiming at the problem of parameter redundancy of YOLOv8n detection head, parameter sharing and DBB module are used to reduce the number of model parameters and ensure the detection accuracy of the model. The detection algorithm was tested on the expanded SIXary dataset. The experimental results show that the mAP value of this algorithm can reach 93.1, which is 2.9% higher than that of the original model, while the number of parameters is decreased by 8.5%, maintaining the detection speed of original model. The performance of this algorithm is significant improved compared to the original algorithm, verifying the effectiveness of its improvement.