关键词: 多元岭回归, 局部纹理特征, 麦苗密度, K-means聚类, 颜色空间

Abstract: Planting density has a great impact on wheat yield, and reasonable planting density is of great significance for increasing wheat yield. Therefore, it is particularly critical to quickly and accurately estimate the wheat planting density.  This paper proposed a method for estimating wheat planting density based on multi-ridge regression. Firstly, the method transfered the acquired color image of  wheat seedling from RGB to Lab, and used the improved K-means clustering algorithm to segment the wheat seedling in the Lab color space. Secondly, 50 images of known seedling density were selected as training samples and normalized, and then the standard difference the area of the image features and contour features and LBP texture features was calaulated. The 3 characteristic parameters and known features were obtained. As training input, the density of wheat seedlings was used to obtain the functional mapping relationship between wheat seedling density and characteristic parameters. Finally, the image of the wheat seedling to be tested was processed according to the above steps, and 3 characteristic parameters of the test wheat seedling image were obtained as inputs, and substituted into the mapping function to obtain the density value of the tested wheat seedling image. The experimental results showed that when estimating the planting density of 4 different varieties of wheat, the average accuracy of wheat seedling density in single image of field wheat seedlings was 93.99%, and the average relative error was 6.01%. Compared with the existing methods of estimating wheat planting density, the estimating accuracy of this method was significantly improved.

Key words: multivariate ridge regression, local texture features, density of wheat seedling, K-means clustering, color space