Evaluation on uncertainty in agricultural remote sensing classification based on exponential hybrid entropy model in parametric form

Abstract: Uncertainty is the most important factor which affects the quality of remote sensing image classification (RSIC), research on uncertainty in RSIC is a cutting-edge, hot topic in remote sensing application study. Study of RSIC gradually developed from simple qualitative and non-positioning research into specific quantitative and positioning research. At present, a RSIC uncertainty evaluation model based on pixel scale and independent of the classification method should be established. In recent years, some scholars began to use hybrid entropy model to evaluate uncertainty in RSIC. However, these studies did not focus on a particular area and find out a suitable entropy function. How to find out a suitable entropy function which better integrate both fuzziness and randomness and facilitate a wider range of entropy values has always been a difficult point of research. From the discussion above, this paper established a method for evaluating uncertainty in agricultural RSIC based on exponential hybrid entropy in parametric form (EHEP). In this study, firstly, the exponential hybrid entropy function was deduced in parametric form, and EHEP was obtained. EHEP is improvement of hybrid entropy which has the shortcoming of lacking adjustable parameters. After adjusting parameters, entropy function can better integrate fuzziness and randomness and facilitate a wider range of entropy values, so this function is suitable for evaluating RSIC uncertainty. Moreover, by the research on the relationship between the parameters and the entropy function surface, the paper ascertained parameters which are suitable for evaluating uncertainty in farming area RSIC. Secondly, EHEP was used to establish a RSIC uncertainty evaluation model based on pixel scale and independent of the classification method, in order to offer elicitation to simulation of the uncertainty transferred in space model, and to help fill a vacancy in uncertainty evaluation model based on pixel scale and independent of the classification method. Lastly, the EHEP model was used to test and verify in SPOT-5 image of Zhenlai County, Jilin Province. The results indicate that in EHEP when parameters are equivalent to 4 and 1, respectively, the function better integrates fuzziness with randomness, and increases entropy value range by 2.11 times compared with logarithmic hybrid entropy function. In addition, the EHEP model evaluates contribution of different pixels to the uncertainty based on pixel scale and independent of the classification method, and corrects deficiency of error matrix in evaluation of RSIC accuracy. Furthermore, it visually expresses the uncertainty, contributes to the overall mastery of RSIC uncertainty's value, distribution, spatial structure and trend, and locates the coordinates of the area where uncertainty exists. Therefore, the EHEP model can make more accurate expression of the uncertainty in agricultural RSIC with relatively complex objects based on pixel scale and independent of the classification method, effectively bolstering precision of crop planting area extraction and remote sensing-based regional yield estimation.