对MobileNetV3模块部分进行预训练，提高最终模型的训练效果缩短

对MobileNetV3模块部分进行预训练，提高最终模型的训练效果缩短训练时间。实验结果表明经过网络轻量化改进后LYOLO算法模型整体的网络参数量下降至44.74MB，根据混淆矩阵统计数据分布，LYOLO网络平均检测精度达到93.6%，单张图片检测耗时为0.01s，LYOLO整体参数量为YOLOv4的18%，随着IoU的增加，模型也仍然具有较高的mAP，其得到的预测框与真实目标框的重合度更高，目标框定位也更准确。算法整体的检测实时性、综合检测定位性能均获得较大提升。

Pre train the mobilenetv3 module to improve the training effect of the final model and shorten the training time. The experimental results show that after the network lightweight improvement, the overall network parameters of the lyolo algorithm model are reduced to 44.74mb. According to the statistical data distribution of the confusion matrix, the average detection accuracy of the lyolo network is 93.6%, the detection time of a single picture is 0.01s, and the overall parameters of lyolo are 18% of that of yov4. With the increase of IOU, the model still has a high map, and the coincidence between the prediction frame and the real target frame is higher, Target box positioning is also more accurate. The overall detection real-time and comprehensive detection and positioning performance of the algorithm have been greatly improved.

Pre-train the Mobilene ETV 3 module to improve the training effect of the final model and shorten the training time. The experimental results show that after the improvement of network lightweight, the overall network parameters of the LYOLO algorithm model are reduced to 44.74MB. According to the statistical data distribution of confusion matrix, the average detection accuracy of LYOLO network reaches 93.6%, the detection time of a single picture is 0.01s, and the overall parameters of LYOLO are 18% of that of YOLOv4. With the increase of IoU, the model still has a high mAP, and the coincidence degree between the predicted frame and the real target frame is higher, and the target frame is also located. The overall real-time detection and comprehensive detection and positioning performance of the algorithm have been greatly improved.