Inspired by the importance of distance metrics and the structure-preserving ability of features, a novel recognition method for underwater targets, called generalized robust principal component analysis (GRPCA), is proposed in this paper. Several advantages of GRPCA are summarized as follows. First, GRPCA employs the ${l}_{\boldsymbol{ 2,p}} $ -norm as the distance metric for calculating the reconstruction error and variance of projected data and attempts to minimize the sum of the ratios between the reconstruction error and the variance for each data sample. This approach allows it to extract the feature information of an image more accurately, which is important for recognition and representation. Second, the proposed GRPCA algorithm not only is robust but also retains the desirable properties of PCA, such as rotational invariance. Moreover, we present a simple yet efficient iterative update algorithm to solve the GRPCA problem. Finally, on the basis of GRPCA, underwater target recognition technology is developed. The extensive experiments on several underwater optical image databases show that our method is more effective and advantageous than other subspace learning algorithms are.
26 June 2023 in IEEE Transactions on Circuits and Systems for Video Technology
Currently, many robust principal component analysis (PCA) methods with low-dimensional representations have been proposed to improve the overall performance of image recognition. However, most existing methods are still sensitive to noise in the selection of structure information. To overcome this problem, we propose a novel structure for PCA, called arbitrary triangle structure adaptive mean PCA (ATAM-PCA). On the basis of ensuring that the variance, reconstruction error and input data with the flexible l 2, p -norm can establish a triangle structure, ATAM-PCA maximizes the summation of the difference between the variance and reconstruction error of each projection data sample, which improves the universality and robustness of the overall structure, but also successfully protects the rotational invariance and geometry of the data. Moreover, ATAM-PCA applies the adaptive mean strategy to perform the data centralization task, which further effectively reduces the negative impact of noise. Finally, we design a fast iterative algorithm for solving ATAM-PCA. Extensive experimental results based on several image databases show the effectiveness and advantages of our method.
Recently, many PCA with robust low-dimensional representation models have been applied in imaging. However, most models ignore the manifold geometry of the data and fail to minimize the reconstruction error. Here, a novel robust PCA structure, locally invariant robust principal component analysis (LIRPCA), is proposed for underwater image recognition. The contributions of LIRPCA are as follows: (1) LIRPCA selects the ℓ 2 -norm as distance metric criterion to describe the global geometry and the intrinsic geometry, which ensures the robustness of the overall model structure. (2) LIRPCA constructs a close relationship between the reconstruction error of the projected data and the input data in the cost function to minimize the reconstruction error. (3) To solve the challenging optimization function of LIRPCA, we design an iterative algorithm with fast convergence to obtain the desired solution. The proposed model is applied to feature extraction and recognition tasks from several underwater image datasets and performs better than other models.
This paper studies an advanced intelligent recognition method of underwater target based on unmanned underwater vehicle (UUV) vision system. This method is called kernel two-dimensional nonnegative matrix factorization (K2DNMF) which can further improve underwater operation capability of the UUV vision system. Our contributions can be summarized as follows: (1) K2DNMF intends to use the kernel method for the matrix factorization both on the column and row directions of the two-dimensional image data in order to transform the original low-dimensional space with nonlinearity into a higher dimensional space with linearity; (2) In the K2DNMF method, a good subspace approximation to the original data can be obtained by the orthogonal constraint on column basis matrix and row basis matrix; (3) The column basis matrix and row basis matrix can extract the feature information of underwater target images, and an effective classifier is designed to perform underwater target recognition; (4) A series of related experiments were performed on three sets of test samples collected by the UUV vision system, the experimental results demonstrate that K2DNMF has higher overall target detection accuracy than the traditional underwater target recognition methods.