关键词: 麦穗计数, Bayes抠图算法, 麦穗小穗计数, 图像分割, 产量预测, 连通区域

Abstract: For yield evaluation, the number of ear and spikelet in the field unit area is acquired manually, which often takes time and energy. In order to realize efficient and automatic spikelet counting, this paper proposed an automatic spikelet counting method based on improved Bayes image matting algorithm. Firstly, the improved Bayes matting algorithm was used to mat the wheat ear image under the natural growth condition of the acquired land to separate wheat ear from the natural background. Then the image was smoothed and binarized, and the iterative limit corrosion operation was used to corrode the binarized image, remove the awn in the wheat ear image, and separate each individual wheat ear and spikelets on the wheat ear. The area filter was applied to remove the area that was too small, and the black hole of the remaining area was filled, so that each individual wheat ear spikelet formed a separate connected region. Finally, the connected region was marked and counted, and the automatic counting of wheat ear spikelet was completed. In this paper, the spikelets of four wheat varieties were counted and validated by using the image of spikelets. The experimental results showed that the average counting accuracy of spikelets in single image of four wheat varieties was 94.53% and the average relative error was 5.47%. Compared with the automatic counting method of spikelets in the literature, the counting accuracy of spikelets of four wheat varieties was significantly improved.

Key words: wheat ear counting, improved bayes matting method, grain number counting, image segmentation, yield prediction