Мінімізація кількості інформативних ознак при побудові класифікатора рослинних об’єктів

Abstract

UK: В статті показана можливість зменшення кількості інформативних ознак з початкових 256 значень до 4 при побудові класифікатора рослинних об’єктів за результатами дистанційного зондування. Для побудови розпізнавальної моделі був використаний множинний лінійний дискримінантний аналіз Фішера. Для обраного набору даних, встановлені оптимальні діапазони ознаки. EN: In the article a possibility is considered as to decrease information features at discrimination of vegetable objects. The multiple discriminant Fisher analysis with step-by-step inclusion of variables was used as an algorithm of discrimination. The information features were selected on the basis of the subjectively individual approach. The features were selected from frequency areas in which there existed the greatest difference by a reflection coefficient value between spectral reflection coefficients for each kind of plants. The special-purpose literature on these themes was also taken into account. For the analysis of the discrimination efficiency by the classifier of plants (for a different amount of features) a probability of plant proper discrimination was used. A minimum admissible level of the correct detection probability for each kind was set as 90 %. For investigation there were used the real reflection coefficients of plants – maize, bristlegrass and ambrosia that were measured in field conditions.