Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2019-01-28 Cooperative journals: 《计算机应用研究》
Abstract： Aimed at the problems of hyperspectral unmixing, it was found that the unmixing accuracy of the traditional unmixing algorithm was not high when the number of endmember was kept constant in unmixing. Thus, based on the artificial neural network (ANN) , this paper proposes a novel unmixing algorithm of estimating the number and category of endmember in a single pixel. Firstly, the unmixing algorithm uses the artificial neural network to estimate the number and category of each mixed pixel’s endmember in the remote sensing image. Then, the objective function of the algorithm is determined based on the estimation results, and the improved differential search algorithm is introduced to solve the objective function. Finally, the abundances and the parameters are obtained to realize the unmixing of hyperspectral images. The experimental results on simulated and real hyperspectral data demonstrate that compared with the existing unmixing algorithms, the proposed unmixing algorithm has higher performance and is more in line with the actual scene.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2019-01-28 Cooperative journals: 《计算机应用研究》
Abstract： Some methods for points cloud registration only use the swarm intelligence optimization algorithm and the points’ spatial information. Those methods have such shortcomings as slow convergence speed and long time consuming. To overcome these weakness, this paper proposed a registration algorithm based on the curvature information of points cloud. The algorithm extracted feature points according to the curvature information. And it obtained the best transformation matrix to make the two point cloud coincide by the improved artificial bee colony algorithm. In the process of population optimization, the corresponding points were searched according to the curvature information, and the scale of point cloud was reduced. The experimental results show that compared with the registration algorithm only using random point selection method and point cloud spatial coordinate information, the proposed algorithm can effectively accelerate the convergence speed and significantly shorten the registration time without reducing the registration accuracy.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-20 Cooperative journals: 《计算机应用研究》
Abstract： Traditional evolutionary point cloud registration methods often not using the color information in the models. To overcome the defect, this paper introduced a point cloud registration method based on self-adaptive evolutionary optimization algorithm and color information. It subsampled the input point clouds by extracting the color feature points and randomly chosen points, it utilized the median of all pairs of color constrained points as the object function. At last, it used the self-adaptive evolutionary optimization algorithm to get optimal solution. The registration experiments on four colorized point clouds show that, compared with the evolutionary point cloud registration methods only spatial information use in and two state-of-the-art registration methods, the method significantly shorten the processing time while achieving similar registration precision.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-04-17 Cooperative journals: 《计算机应用研究》
Abstract： According to the slow convergence and easiness to trap in local optimum of backtracking search optimization algorithm, this paper presented an improved algorithm. The method proposed a mutation scale factor based on t distribution firstly to speed up the convergence rate; then the algorithm improved the structure of crossover equation and introduced the optimal individual to control the direction of population search, which effectively improved the development capability. Finally, the algorithm proposed the evolutionary selection mechanism, introduced the mutation factor of differential evolution algorithm and replaced the optimal solution with worse solution under a certain probability, which can avoid algorithm to fall into the local optimum. The numerical experiments selected 15 test functions for simulation and compared with 5 well-behaved algorithms. The results show that the proposed algorithm has obvious advantages in terms of convergence rate and search accuracy.
Hits
Hits
Downloads
Comment