The Aircraft Skin Crack Inspection Based on Different-Source Sensors and Support Vector Machines

Currently, most aircraft skin crack inspections are done by naked-eye vision and video instrumentation, but a single vision inspection can only provide partial information and incomplete description of the crack characteristics. A multi-class classification method for skin crack inspection based on multi-class support vector machines and data fusion strategy is proposed in this paper. A mobile platform for aircraft skin inspection carries the image sensor and ultrasonic sensor which are used to collect the signals of the skin inspection in the information fusion system and the sample features from the two different data sources are centralized to construct a feature input space for the skin crack classification. Then amulti-class support vector machine classifier is trained by using some samples of the data and the parameters of support vector machine (SVM) are optimized by using Genetic Algorithm (GA). The other samples of the data are used to forecast the skin crack. The experimental results showed that the method of the different-source information fusion can improve the aircraft skin crack distinguishing degree and accuracy compared with a single sensor and the GA-SVM method can be a more accurate model to recognize the skin crack. The ultrasonic sensor provides some additional information which a single image sensor and eyes can't find. The recognition accuracy has been effectively improved by using different-source sensor information fusion which overcomes the low resolution and the impact of human factors.