基于轻量级网络和数据扩增的作物与杂草识别

Crop and weed identification based on lightweight network and data amplification

为解决传统的深度学习网络在进行农作物与杂草识别时存在的问题,如训练时间长、识别精度低、检测速度慢、采集数据需求量大等,基于YOLOv4算法设计一种改进检测模型。利用轻量级特征提取网络替代原YOLOv4特征提取网络,在增强特征提取网络引入深度可分离卷积,降低网络参数规模。通过数据扩增方法对原始实验数据进行扩充,增加实验数据量和数据多样性,提高模型识别准确率。实验结果表明,改进模型检测速度约为54帧/s,是原YOLOv4模型的330%,训练时间为原来的21.8%,对自然环境下玉米及其伴生杂草的识别准确率更高。此方法亦适用于其他作物与杂草识别。

To solve the problems of the traditional deep learning network in identifying crops and weeds, such as long training time, low recognition accuracy, slow detection speed and large demand for data collection, an improved identification model is proposed based YOLOv4 algorithm.The lightweight feature extraction network is used to replace the original YOLOv4 feature extraction network, and the deep separable convolution is introduced into the enhanced feature extraction network to reduce the scale of network parameters.The original experimental data is expanded by data amplification method to increase the number and the diversity of experimental data, and to improve the accuracy of model recognition.The experimental results show that the detection speed of the improved model is 54 fps.Compared with the original YOLOv4 model, the detection speed is 330% and the training time is 21.8%.And the identification accuracy for corn and its associated weeds in the natural environment is higher.This method can be applied to the identification of other crops and weeds.