The Field Wheat Count Based on the Efficientdet Algorithm
Liangben Cao, Xixin Zhang, Jingyu Pu, Siyuan Xu, Xinlin Cai, Zhiyong Li
Abstract
It is of great significance to realize the low cost and fast statistics of wheat ears to predict wheat yield. At present, a variety of methods have been applied to the measurement of wheat planting density, although these methods can count ears of wheat, they are expensive and difficult to be put into actual production. In order to further improve the accuracy of wheat ear identification and detection counting under field environment, based on the image processing and deep learning technology, EfficientDet algorithm is proposed to detect the wheat ear image. The idea is to frame out the wheat in the wheat ear image and then count the detected target number to realize the automatic counting of wheat ears. EfficientDet-D3 model is adopted in this paper, and adamw algorithm is used to train the model, with a learning rate of 1e5. The experimental results show that the model can quickly and accurately recognize wheat ear images with different densities under various lighting conditions. The final accuracy rate reaches 92.92%, and the test time of the single sheet is 0.2s. We carry out ear counting test on 20 wheat ear images with different densities, and the average accuracy rate reaches 95.30%. This method can detect wheat ear image in a very short time and the detection effect is good. To some extent, the accuracy of wheat ear counting has been increased, and has the potential to be put into actual production.