The Classification of Farming Progress in Rice-Wheat Rotation Fields Based on UAV RGB Images and the Regional Mean Model

Extraction of farming progress information in rice-wheat rotation regions is an important topic in smart field research. In this study, a new method for the classification of farming progress types using unmanned aerial vehicle (UAV) RGB images and the proposed regional mean (RM) model is presented. First, RGB information was extracted from the images to create and select the optimal color indices. After index classification, we compared the brightness reflection of the corresponding grayscale map, the classification interval, and the standard deviation of each farming progress type. These comparisons showed that the optimal classification color indices were the normalized red-blue difference index (NRBDI), the normalized green-blue difference index (NGBDI), and the modified red-blue difference index (MRBDI). Second, the RM model was built according to the whole-field farming progress classification requirements to achieve the final classification. We verified the model accuracy, and the Kappa coefficients obtained by combining the NRBDI, NGBDI, and MRBDI with the RM model were 0.86, 0.82, and 0.88, respectively. The proposed method was then applied to predict UAV RGB images of unharvested wheat, harvested wheat, and tilled and irrigated fields. The results were compared with those obtained with traditional machine learning methods, that is, the support vector machine, maximum likelihood classification, and random forest methods. The NRBDI, NGBDI, and MRBDI were combined with the RM model to monitor farming progress of ground truth ROIs, and the Kappa coefficients obtained were 0.9134, 0.8738, and 0.9179, respectively, while traditional machine learning methods all produced a Kappa coefficient less than 0.7. The results indicate a significantly higher accuracy of the proposed method than those of the traditional machine learning classification methods for the identification of farming progress type. The proposed work provides an important reference for the application of UAV to the field classification of progress types.

Automated summary

In this study, a new method for the classification of farming progress types using unmanned aerial vehicle (UAV) RGB images and the proposed regional mean (RM) model is presented. The method combines optimal color indices and the RM model to improve classification accuracy. Experimental results demonstrate that the proposed method achieves higher accuracy in identifying farming progress types compared to traditional machine learning methods.