Detection of canopy water content of winter wheat during wintering period based on image features

In order to accurately and easily determine the canopy moisture content of winter wheat during wintering period, methods of image processing and feature application based on visible light image were researched. According to the illumination invariance and color constancy principle, the combinational algorithm of homomorphic filtering and multi-scale Retinex was proposed for illumination enhancement processing to eliminate the adverse effects of natural light condition. Totally 39 initial image features which belonged to color, texture and morphology were extracted and investigated, remarkable features selection was conducted by correlation analysis and hypothesis testing. Partial least squares regression was then adopted to establish the water content detection model of canopy. Test results for two winter wheat varieties of “Huai-mai 30” and “Yan-nong 19” showed that the mean relative error and variance of the proposed method were 1.290% and 1.053, respectively, which had no obvious differences between the two varieties, and the detection errors were slightly large in sunny days and noon. The results indicated that the proposed method had high detection accuracy and good adaptability. The key issues of the field image enhancement and image feature selection were studied, and the results are helpful to improve the practicability of crop moisture detection based on the computer vision technology under the background of agricultural internet of things. Meanwhile, the canopy moisture content detection model of winter wheat during wintering period which was established based on this method has good performance, and it can provide effective technical support for winter wheat freeze-proofing and drought resistant decision.