Método híbrido Kendall–Kolmogorov–Smirnov para la selección de características orientada a la evaluación automática de la calidad del durazno

Resumen

The present work presents a hybrid methodology for the selection of traits in peach samples from the municipality of Atltzayanca, (Tlaxcala). The methodology consists of the application of Kendall's Tau coefficient, which measures the correlation by ranges between a variable and the class. This allows you to detect if, when the value of a feature increases, the class also changes in an orderly fashion (high values belong to one class and low values to another). This method is useful for identifying monotonic relationships that are not necessarily linear. In a complementary way, the Kolmogorov–Smirnov test is used, which compares the distributions of the values of the same variable between two classes; If these distributions are different with a significance level p = 0.05, the attribute is considered to clearly distinguish between the two classes. Feature extraction is done using trained YOLO models, accurately trimming the fruits to measure their attributes. The hybrid method is applied to the entire set of images as a subset stabilized by bubble sorting; in both cases, results close to 100% are obtained, which demonstrates the high capacity of each characteristic to separate peaches of acceptable quality (CA) from those of unacceptable quality (NA). By using the hybrid method, not only correlation and distributions are considered, but the characteristic is also required to be consistent in both directions, reducing the risk of selecting misleading variables and allowing the selected characteristics to be analytically justified.