Deep Learning-Based Crop Weed Recognition

To achieve accurate weed control, variable spraying of herbicides and accurate identification of weeds is a prerequisite. In this study, the research on weed recognition is carried out by using deep learning techniques with crop weed images obtained from complex backgrounds in the natural environment of crop production. Firstly, the original image is preprocessed to eliminate irrelevant information in the image, recover useful and real information, enhance the detectability of relevant information and maximally simplify the data; secondly, the crop weed recognition model based on VGG-16 convolutional neural network is constructed, and its convolutional layer is locally adjusted to optimize the main model parameters, so as to achieve the effective targeting of crop weed recognition. Finally, the techniques, methods and models used in this study can be applied to weed identification in complex backgrounds, which can provide technical support for crop herbicide targeting on the basis of weed species identification in this project, as well as provide references and reference for the research of species identification of companion weeds in other crops.