Extração automática de características de asas de mosca da espécie Drosophila melanogaster

Abstract

Biometrics techniques are used in studies of several animal species, such as the Drosophila melanogaster, popularly known as fruit fly. This species has become a model organism for the study of the impacts that other insects produce to the environment. In addition, these flies have proteins and genes similar to those of humans. A challenge for specialists in the study of these individuals is the similarity of the wings between males and females, which may be affected by mutations or variations in the genotype. This work proposes a method of discrimination for gender and genotype of flies of the species Drosophila melanogaster from features extracted from images of the wings. This approach is based on the fractal dimension extracted from the Canny filter segmentation of the components of the Stationary Wavelet Transform. The methodology is validated by dividing the data into groups with variable training and test rates and using six different classifiers: Random Forest, Support Vector Machines, Multi-layer Perceptron, Linear Discriminant Analysis, Quadratic Discriminant Analysis and K Nearest Neighbors. Then, the classification is repeated with data reduction, only females for genotype or only individuals without mutation for gender. The results were satisfactory, surpassing works of the literature with automatic methodologies. For genotype, the hit rates were lower due to the physical similarity between the wings.