THE NEW METHOD FOR ACCURACY ASSESSMENT OF IMAGE CLASSIFICATION OBTAINED USING UNMANNED AERIAL VEHICLES BASED ON A WEIGHTED CONFUSION MATRIX AND ITS ACCURACY COEFFICIENTS

The proposed new method for accuracy assessment of image classification in UAV-based Remote Sensing can be applied in solution of different ecological and practical tasks. Nowadays thematic maps play an important role in solution of different remote sensing tasks. Thematic maps are applied for forest classification, determing of soil types and properties, environmental monitoring, exploring of oil and gas. That’s why the accuracy assessment is necessary to evaluate the quality of thematic maps. It is important to know the accuracy of thematic maps before they are used for further scientific investigations. Users and producers of maps compare several maps to see which is best, or to check how well they agree. It was proposed to use Weighted confusion matrix for accuracy assessment of thematic maps. Proposed Weighted confusion matrix was considered with Confusion matrix. It was noted, that Confusion matrix needs in large samples and can not take into account the “seriousness” of errors. It also were shown main advantages of Weighted confusion matrix. It was noted, that Weighted confusion matrix gives different weights for different mistakes of classification. Proposed Weighted confusion matrix gives a partial credit for classification results. This property of the Weighted confusion matrix is very important, when not all mistakes are equally serious and rough for user. Proposed method uses the Weights matrix for Confusion matrix that contains weights for each element in the Confusion matrix. Accuracy coefficient of the Weighted confusion matrix, such as: Overall accuracy, User’s accuracy, Producer’s accuracy and Weighted average of the weights for each class and their main properties were described in this work too. It was also considered a numerical example of calculation of accuracy coefficients of Weighted confusion matrix. This proposed new method for accuracy assessment of image classification can be applied in land-cover classification, environmental monitoring, exploring for minerals, numerous agricultural tasks.