Artificial bee colony algorithm is an effective algorithm for parameter optimization, but the traditional artificial bee colony algorithm is liable to fall into local extreme points at a later stage. In this paper, we propose an improved artificial bee colony optimization algorithm, which solves the problems of premature convergence and falling into the local extreme value in the classification of hyperspectral images. First we use an improved chaotic sequence with higher randomness to initialize and update nectar sources to expand the distribution of nectar sources. Secondly, the optimized adaptive step size is introduced into the neighborhood search to speed up the algorithm convergence and improve the search efficiency. Then we build an improved artificial bee colony algorithm support vector machine optimization model to mine the optimal values of penalty factor C and kernel function parameter σ. Next, the model was used to perform classification experiments on two hyperspectral images (University of Pavia, Indian Pine) with different attributes, and compared with the traditional bee colony algorithm, genetic algorithm, and particle swarm algorithm. Experimental results on HSI datasets demonstrate the superiority of the proposed method over several well-known methods in both classification accuracy and convergence speed. INDEX TERMS Artificial bee colony, support vector machine, hyperspectral image, chaotic sequence.

At present, the classification of the hyperspectral image (HSI) based on the deep convolutional network has made great progress. Due to the high dimensionality of spectral features, limited samples of ground truth, and high nonlinearity of hyperspectral data, effective classification of HSI based on deep convolutional neural networks is still difficult. This paper proposes a novel deep convolutional network structure, namely, a hybrid depth-separable residual network, for HSI classification, called HDSRN. The HDSRN model organically combines 3D CNN, 2D CNN, multiresidual network ROR, and depth-separable convolutions to extract deeper abstract features. On the one hand, due to the addition of multiresidual structures and skip connections, this model can alleviate the problem of over fitting, help the backpropagation of gradients, and extract features more fully. On the other hand, the depth-separable convolutions are used to learn the spatial feature, which reduces the computational cost and alleviates the decline in accuracy. Extensive experiments on the popular HSI benchmark datasets show that the performance of the proposed network is better than that of the existing prevalent methods.

No abstract

No abstract

No abstract