Image Augmentation For Aircraft Parts Detection Using Mask R-CNN

Aircraft maintenance is an endeavor to ensure flight safety. Visual inspection is a component of aircraft maintenance that seeks to ascertain the condition of crucial aircraft parts, including the fuselage, wings, tail, engines, and doors. This research aims to implement image augmentation for aircraft part detection. Augmentation is used to increase the dataset to obtain more accurate detection results, as the currently available dataset is very limited. The augmentation techniques used include four groups: transform, pixel, noise, and a combination of the three. For the detection of aircraft parts, we use the Mask R-CNN model, which has three main stages: (i) convolution stage for feature extraction, (ii)region proposal stage to generate object proposals, and (iii) mask prediction stage to generate object segmentation masks. Based on the experiment, the highest accuracy is obtained in the transformation augmentation group, with a value of 90.02%. The experimental results indicated that the proposed approach achieved a significant level of accuracy in detecting aircraft parts.