Application Research Of Classification Method Of Combination Technology

Fire to humans and property caused serious harm, predict fire timely can avoid losses.At present, the influence factors of detection equipment pop ular susceptible to environmental, weather, which restricted to the application of traditional detection equipment. Smoke is ahead of fire, and smoke propagation area is large and not easy to be occluded objects, fire in the early time on smoke detection can effectively prevent fire occurrence. In recent years, with the development of artificial intelligence and the digital image processing technology, video smoke detection technology has achieved great progress. But in reality, the smoke detection technolo gy still has many problems such as classification method is easily affected by the noise and environmental impact, resolution and time efficiency is not ideal or inspection method is short of adaptability etc.In this paper,we mainly studied the following research of three aspects on artificial intelligence and digital image processing:1. At present the classical classification algorithms have been developed to bottleneck, by combining different classification algorithms can effectively improve the precis ion of classification. For finite samples, KNN heavily depends on an appropriate distance while distance learning on previous studies didn’t fully consider the distribution of the sample, in this paper, a new probability-based two- level nearest neighbor classification algorithm for adaptive metrics(PTLNN) is proposed. This proposed algorithm is divided into two levels, uses Euclidean distance in the low- level to determine an unlabeled sample local subspace set; at the high- level, using Ada Boost extracts information from subspace. Then to minimize the mean absolute error of principle, defines a probability-based adaptive distance nearest neighbor classifier. The proposed algorithm PTLNN combines the advantages of KNN and Ada Boost algorithm, fully considers the sample distributions in finite sample to reduce error rate, and has good stability in noisy data to reduce Ada Boost’ s overfitting phenomenon. In contrast to other algorithms, experimental results show PTLNN can achieve better results.2. A probability-based two- level nearest neighbor classification algorithm(PTLNN) is proposed to detect smoke in video. The proposed algorithm adopts the discrete cosine transform(DC T) and discrete wavelet transform(DWT) in two ways to extract smoke characteristics. By comparing with the traditional algorithms, the proposed PTLNN algorithm with the discrete cosine transform has better effectiveness on video smoke detection in not only meeting the real-time requirements but also improving the detection accuracy.3. For finite samples, KNN heavily depends on an appropriate distance while the generalization is not strong and the time efficiency is low for previous studies of distance learning. In this paper, a new sparsity- inspired two- level classification algorithm(STLCA) is proposed. This proposed algorithm is divided into two levels, uses Euclidean distance in the low- level to determine an unlabeled sample local subspace set; at the high- level, using sparse Bayesian extracts information from subspace. Because of its spars ity, in noisy situations STLCA can have good stability and strong generalization, and high time efficiency. Through the noise data and applications in video smoke detection indicates, STLC A algorithm can achieve better results.