摘要
传统图像识别率及识别准确率不高,检测速度比较慢,在计算能力比较低的设备中无法运行等,对此使用改进的Tiny-YOLOv3算法模型解决这些问题。改进的算法模型进一步减少了模型的尺寸,检测速度大大提升,检测准确率与原模型的相比并没有太大变化。实验结果表明,改进的Tiny-YOLOv3算法生成的模型的尺寸为8.5 MB,比原模型更小,同时在数据集上的实时性能表现为25.3 FPS,mAP值为60%左右,性能比原模型更优。
The traditional image recognition rate is not high,the recognition accuracy rate is relatively low,the detection speed is relatively slow,and can not run in devices with relatively low computing power,etc.Aiming at these problems,use an improved Tiny-YOLOv3 algorithm model to solve these problems.The improved algorithm model further reduces the size of the model,and the detection speed is also greatly improved,but the detection accuracy has not changed much from the original model.The experimental results show that the size of the model generated by the improved Tiny-YOLOv3 algorithm is 8.5 MB,which is smaller than the original model.Meanwhile,the real-time performance on the data set is 25.3 FPS,the mAP value is about 60%,and the performance is better than the original model.
作者
王少韩
刘淼
Wang Shaohan;Liu Miao(School of Mechanical and Automotive Engineering,Shanghai University of Engineering Technology,Shanghai 201620,China)
出处
《农业装备与车辆工程》
2022年第9期121-124,共4页
Agricultural Equipment & Vehicle Engineering