We propose a derivative-free trust region algorithm with a nonmonotone filter technique for bound constrained optimization. The derivative-free strategy is applied for special minimization functions in which derivatives are not all available. A nonmonotone filter technique ensures not only the trust region feature but also the global convergence under reasonable assumptions. Numerical experiments demonstrate that the new algorithm is effective for bound constrained optimization. Locally, optimal parameters with respect to overall computational time on a set of test problems are identified. The performance of the best choice of parameter values obtained by the algorithm we presented which differs from traditionally used values indicates that the algorithm proposed in this paper has a certain advantage for the nondifferentiable optimization problems.
In this paper, we design an end-to-end adaptive feature normalization network (AFN-Net) for single image dehazing. In order to fit the function that can recover haze-free images from haze images more effectively, an Adaptive Normalization module (AN) is designed, which can unify the input into the same feature space. A large number of experimental evaluations show that the proposed method is superior to the state-of-the-art (SOTA) method on the benchmark datasets.
In this paper, a new three-term conjugate gradient algorithm is proposed to solve unconstrained optimization including regression problems. We minimize the distance between the search direction matrix and the self-scaling memoryless BFGS direction matrix in the Frobenius norm to determine the search direction, which has the same advantages as the quasi-Newton method. At the same time, random parameter is used so that the search direction satisfies sufficient descent condition. For uniformly convex functions and general nonlinear functions, we all establish the global convergence of the new method. Numerical experiments show that our method has nice numerical performance for solving large-scale unconstrained optimization. In addition, the application of the new method to the regression model proves that our method is effective.
Hyperspectral images (HSIs) will experience noise throughout the data collection process due to the imaging system's limitations, which will make it challenging to extract the image's crucial information. In this paper, a multi-stage enhanced HSI denoising network (MED-Net) is proposed. Our core concept is to process the hyperspectral noise image iteratively using a multi-stage network. A similar network structure's first and second phases are employed for the denoise process. To achieve cross-stage information transfer, we use CSFF (Cross-stage Feature Fusion) mechanism and SAM (Supervised Attention Module). AN (Additive Network) and MN (Multiplicative Network) are used to remove additive and multiplicative noise. Then, we restore the background based on the residual network and attention mechanism. The results of our experiments demonstrate the superiority of our approach over the actual HSIs data recovery, and the restored image has good visual clarity and detail.
The new nonsingularity criteria of matrices and the equivalent representation of M-matrices are presented in this paper. The recent relevant results are extended.
'/%3e%3cuse xlink:href='%23a' transform='rotate(23.036 .093 25.536)'/%3e%3cuse xlink:href='%23a' transform='rotate(45.87 1.273 16.18)'/%3e%3cuse xlink:href='%23a' transform='rotate(69.945 .996 12.078)'/%3e%3cuse xlink:href='%23a' transform='rotate(20.66 -19.689 31.932)'/%3e%3c/svg%3e)