Prediction and counting of field wheat based on LC-DcVgg

The number of wheat spikes per unit area is an important parameter for assessing wheat yield and wheat planting density. At present, the methods of intelligent counting of wheat include remote sensing technology and machine learning technology, but all have shortcomings such as poor stability, strong limitations, and poor versatility. And however, the existing object detection neural network algorithm requires a large amount of manpower to produce data sets. And it is not possible to identify too dense wheat spikes. In this paper, a new structure called direct connection is proposed to improve the algorithm of Vggnet, which makes it better combined with localization-based counting loss. Direct connection can fuse the features of shallow layer with those of deep layers, which can make the network retain the original picture information and make the localization-based counting loss play a better role in wheat spike counting. The model has a good recognition effect on the wheat spikes that are stuck together. For dense wheat spikes, the model can achieve MAE of 11.857, an accuracy rate of 90.4%, RMSE of 16.985.