Buckwheat Grain Image Segmentation Based on U-Net Model

Abstract

In order to improve the segmentation effect of buckwheat grain image, the improved U-Net network was used to increase the network depth and the Batch Normalization layer was added; Transfer learning is introduced and Pascal VOC pre training model is adopted to improve the performance of the model; A new loss function combining Dice loss and Cross Entropy loss (CEloss) is used to mitigate the sample imbalance. The buckwheat grain mixture images collected by the image acquisition experimental platform were annotated, 175 512 pixel images were selected as the initial samples, and data enhancement was used to expand the data to improve the robustness and generalization ability of the model. The test results showed that the comprehensive evaluation index value of the segmentation of non shelled buckwheat was 97. 99%, the comprehensive evaluation index value of the segmentation of intact buckwheat rice was 89. 62%, and the comprehensive evaluation index value of the segmentation of broken buckwheat rice was 72. 70%. The buckwheat grain image segmentation algorithm based on the U-Net model proposed in this paper can effectively segment the unwrapped buckwheat, intact buckwheat rice and broken buckwheat rice in the buckwheat grain image, with higher accuracy, and is of great significance to the realization of adaptive optimal control of the buckwheat husker.