28 November 2019 in IEEE Access
During the past decade, deep learning is one of the essential breakthroughs made in artificial intelligence. In particular, it has achieved great success in image processing. Correspondingly, various applications related to image processing are also promoting the rapid development of deep learning in all aspects of network structure, layer designing, and training tricks. However, the deeper structure makes the back-propagation algorithm more difficult. At the same time, the scale of training images without labels is also rapidly increasing, and class imbalance severely affects the performance of deep learning, these urgently require more novelty deep models and new parallel computing system to more effectively interpret the content of the image and form a suitable analysis mechanism. In this context, this survey provides four deep learning model series, which includes CNN series, GAN series, ELM-RVFL series, and other series, for comprehensive understanding towards the analytical techniques of image processing field, clarify the most important advancements and shed some light on future studies. By further studying the relationship between deep learning and image processing tasks, which can not only help us understand the reasons for the success of deep learning but also inspires new deep models and training methods. More importantly, this survey aims to improve or arouse other researchers to catch a glimpse of the state-of-the-art deep learning methods in the field of image processing and facilitate the applications of these deep learning technologies in their research tasks. Besides, we discuss the open issues and the promising directions of future research in image processing using the new generation of deep learning.
24 June 2016 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
In this paper, a novel polarimetric synthetic aperture radar (PolSAR) image unsupervised classification method is proposed. It combines three typical features, including polarimetric data features (coherent matrix), polarimetric decomposition features (Krogager, Freeman, Yamaguqi, Neuman, and H/A/α decomposition), and gray-level co-occurrence matrix features to comprehensively describe the data characteristics. And it also proposes a symmetric revised Wishart (SRW) distance-derived manifold regularized low-rank representation (SRWM_LRR) method to deeply exploit the geometry data structure. The low-rank representation (LRR) is used to capture the intrinsic global structure of PolSAR data and the manifold regularization is employed to detect the local structure of the data, in which SRW distance is introduced to measure the similarity between different pixels for describing the local manifold structure. This algorithm considers the specific statistics property in PolSAR data and simultaneously integrates multiple features in perspective of data geometry structure to represent pixels for achieving a better classification performance. The effectiveness and practicability of the proposed method are demonstrated by datasets obtained either in spaceborne or airborne SAR system, including the Flevoland dataset (AIRSAR L-Band) extensively used in land classification cover, and Xi'an dataset (RadarSAT-2 C-Band). Compared with the traditional Wishart classifier, Euclidean and SRW distance-based spectral clustering and LRR, the proposed method shows improvement in accuracy and efficiency as well as a better visualization result.
24 June 2019 in IEEE Access
A novel deep architecture, the sparse deep tensor extreme learning machine (SDT-ELM), is presented as a tool for pattern classification. In extending the original ELM, the proposed SDT-ELM gains the theoretical advantage of effectively reducing the number of hidden-layer parameters by using tensor operations, and using a weight tensor to incorporate higher-order statistics of the hidden feature. In addition, the SDT-ELM gains the implementation advantage of enabling the random hidden nodes to be added block by block, with all blocks having the same hidden layer configuration. Moreover, an SDT-ELM without randomness can also achieve better learning accuracy. Extensive experiments with three widely used classification datasets demonstrate that the proposed algorithm achieves better generalization performance.
4 June 2019 in IEEE Access
Fast Sparse Deep Neural Networks that aims to offer an alternative way of learning in a deep structure is proposed in this paper. We examine some optimization algorithms for traditional deep neural networks and find that deep neural networks suffer from a time-consuming training process because of a large number of connecting parameters in layers and layers. To reduce time consumption, we propose Fast Sparse Deep Neural Networks, which mainly considers the following two aspects in the design of the network. One is that the parameter learning at each hidden layer is given utilizing closed-form solutions, which is different from the BP algorithm with iterative updating strategy. Another aspect is that Fast Sparse Deep Neural Networks uses the summation method of a multi-layer linear approximation to estimate the output target, which is a different way from most deep neural network models. Unlike the traditional deep neural networks, Fast Sparse Deep Neural Networks can achieve excellent generalization performance without fine-tuning. In addition, It is worth noting that Fast Sparse Deep Neural Networks can also effectively overcomes the shortcomings of the extreme learning machine and hierarchical extreme learning machine. Compared with existing deep neural networks, enough experimental results on benchmark datasets demonstrate that the proposed model and optimization algorithm are feasible and efficient.