AUTHOR=Wang Yiding , Qin Yuxin , Cui Jiali 

TITLE=Occlusion Robust Wheat Ear Counting Algorithm Based on Deep Learning

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 12 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2021.645899

DOI=10.3389/fpls.2021.645899

ISSN=1664-462X

ABSTRACT=Evaluation of wheat ears yield is of great significance to modern intelligent agriculture. Counting the number of wheat ears in the wheat image under natural light is the main way to evaluate the yield of wheat ears. However, the distribution of wheat ears is dense, thus the occlusion and overlap problem appears in almost every wheat image. It is difficult for traditional image processing methods to solve the problem. In comparison, deep learning networks provide a new solution to this problem. This paper proposes an improved EfficientDet-D0 object detection model for wheat ear counting. First, the transfer learning method is used in the pre-training of the model backbone network to extract the high-level semantic features of wheat ears. Secondly, an image augmentation method Random-Cutout is proposed to erase part of the rectangular area according to the number and size of the wheat ears in the images to simulate occlusion in real wheat images. Finally, the convolutional block attention module (CBAM) is adopted into the EfficientDet-D0 model after the backbone. It makes the model pay more attention to the wheat ears and suppresses other useless background information. With extensive experimental results, the improved EfficientDet-D0 model reaches 94% counting accuracy, which is about 2% higher than the original model. False detection rate is the best among several methods of experimental comparison. Under the best accuracy, the inference time is less than 40ms.