Recently, Kernel Correlation Filter (KCF) is recognized as one of the best visual tracking algorithms which provide excellent tracking performance and high possessing speed. However, how to adapt to the scale change of target in real time and accurately is still an open problem. In this paper, a multi-scale improved method based on one-dimensional scale correlation filtering is proposed. We use the kernel function to solve the regularized least squares classifier, and the translation filter template is obtained to estimate the target position. Then, we collect a series of samples of different scales at the target position to construct a one-dimensional scale filter template for estimating the most suitable scale. The filter is learned and updated online in the tracking process. In scale estimation, we employed the dimension reduction method to reduce the computational complexity, and the interpolation method is used to increase the number of multi-scale to obtain more accurate scale location. A number of experimental results and data show that the target region overlap rate and tracking rate of the improved algorithm are improved by 9% and 60% on average compared with the best of other scale adaptive tracking algorithms. Compared with the KCF algorithm, the average position tracking error is reduced by 23.9% and the overlap rate is increased by 46%.
In the convolutional neural network model, the learning rate represents the magnitude of the neural network parameters that change at each iteration. When the learning rate is too high, the loss function will oscillate without converging. When the learning rate is too low, the loss function converges slowly. How to set the appropriate value of the learning rate becomes an important issue. Based on the learning rate annealing algorithm, this paper sets the segmentation attenuation and adds periodic pulse perturbation. The learning rate gradually declines and rises at the end of the cycle, and then continues to fall. This prevents the network from oscillating at a local minimum or saddle point in the late training period due to the low learning rate . Finally, this paper verifies the method in the application of ethnic clothing classification using the transfer learning VGG-16 model.