Multi-sensor detected object classification method based on support vector machine

Multi-sensor detected data often have noise. The current multiple classification algorithms are susceptible to noise interference, have weak fault-tolerance, and can lead to data misclassification. The multi-sensor detected object classification method was proposed in order to solve the problems. Noise-tolerance least squares projection twin support vector machine (NLSPTSVM) was presented in order to remove outliers to improve noise-tolerance. NLSPTSVM with confidence-degree, based on the defined confidence-degree of NLSPTSVM and the minimal hypersphere distance, was used as binary classifier, and advanced the noise reduction process before the generation of directed graph, according to the idea that "the upper classification performance has more effects on the generalization performance of classification model". A high accuracy, noise-tolerance and fault-tolerance multiple classification support vector machine was proposed, called noise-tolerance up-preferred multiple directed acyclic graph support vector machines (NUMDAG-SVMs). Experiments were conducted to test the performance of the algorithm. Experimental results in public datasets indicate that our NUMDAG-SVMs have comparable classification accuracy, better noise-tolerance and fault-tolerance to those other algorithms. The algorithm can get good classification performance in sensor data.