基于像素直方图和YOLO优化算法的视觉图像识别检测技术研究  

Visual Image Recognition and Detection Technology Based on Pixel Histogram and YOLO Optimization Algorithm

作　　者：张琴[1] ZHANG Qin(Department of Information Engineering,Fuzhou Vocational and Technical College,Fuzhou,Fujian 350108,China)

机构地区：[1]福州职业技术学院信息工程系,福建福州350108

出　　处：《河北北方学院学报（自然科学版）》2023年第3期10-14,共5页Journal of Hebei North University:Natural Science Edition

基　　金：福州职业技术学院引导计划项目“计算机视觉姿态估计技术研究与应用(院士工作站专项)”(FZYKJZXYD202201)。

摘　　要：针对传统识别检测技术对较小目标的检测准确度不足,且检测速度较慢的问题,提出利用小波变换进行图像整合得到原始的背景图像,并引入图像特征融合机制和感受野增强机制优化YOLO算法。对改进像素直方图方法的性能验证结果表明,像素直方图+小波变换检测方法在数据集1~4中的准确率分别为90.39%、94.29%、96.69%和96.60%。在数据集1~4中改进的YOLO算法在目标检测的准确度和检测速度高于YOLOv2和RetinaNet算法。从结果可以看出,改进后的方法具有较高的准确性,可以实现对较小目标的检测,在视觉图像识别技术中可以得到较好的应用,具有较高的应用价值。In view of the problem that the insufficient detection accuracy and the slow detection speed of traditional recognition and detection technology for small targets,this study proposes to use wavelet transform to integrate images to get the original background image,and introduce the image feature fusion mechanism and receptive field enhancement mechanism to optimize YOLO(You Only Look Once)algorithm.The performance verification results of the improved pixel histogram method showed that the accuracy of the pixel histogram+wavelet transform detection method in dataset 1-4 was 90.39%,94.29%,96.69%and 96.60%,respectively.The improved YOLO algorithm in data set 1-4 had higher accuracy and detection speed in target detection than YOLOv2 and RetinaNet algorithms.The results show that the improved method has higher accuracy,achieve the detection of small targets,and can be better applied in visual image recognition technology with higher application value.

关 键 词：像素直方图 YOLO算法 视觉图像识别 检测技术 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]