多尺度融合与FMB改进的YOLOv8异常行为检测方法

Improved YOLOv8 Method for Anomaly Behavior Detection with Multi-Scale Fusion and FMB

针对异常行为检测目标面临多尺度变化、易漏检误检以及复杂背景干扰等问题,提出了一种多尺度特征融合与快速多交叉结构改进的行人异常行为检测方法。该方法以YOLOv8为基线网络,在模型主干部分设计了快速多交叉结构提升上下文信息感知能力并减少网络参数,提出空间递进卷积池化模块实现多尺度信息融合,降低尺度差异带来的易漏检误检问题,提高检测的准确度;在模型颈部中引入SimAM注意力机制抑制复杂背景干扰,提升目标检测性能;最后采用WIoU损失函数平衡检测锚框的惩戒力度增强模型泛化性能。所提方法在UCSD-Ped1、UCSD-Ped2数据集进行验证,并在OPIXray数据集进行了泛化性测试。结果表明,所提方法异常行为识别精度较诸多先进检测算法均有不同程度的提升,且参数量更小,是一种性能较为优异的行人异常行为检测方法。

To resolve the problems of anomaly behavior detection including multi-scale variations,miss and false detec-tion,and complex background interference,a method is proposed by incorporating the fusion of multi-scale features and fast multi-cross block(FMB)for anomaly behavior detection.Based on YOLOv8 as the baseline network,a FMB has been designed in the backbone to enhance context information awareness and reduce network parameters.Meanwhile,a spatial-progressive convolution pooling(S-PCP)module has been proposed to achieve multi-scale information fusion,thereby reducing the issues of miss and false detection caused by scale differences and improving detection accuracy.A SimAM attention mechanism has been introduced in the neck to suppress complex background interference and improve object detection performance.And WIoU has been used to balance the penalty force on anchor boxes,enhancing the model’s generalization performance.The proposed method has been extensively validated on the UCSD-Ped1 and UCSD-Ped2 datasets,and its generalization has been tested on the OPIXray dataset.The results indicate that the proposed method with fewer parameters achieves different improvements in anomaly behavior recognition accuracy compared to many advanced detection algorithms,demonstrating an excellent detection method for pedestrian anomaly behavior.