Search

While achieving remarkable success in remote sensing (RS) scene classification for the past few years, convolutional neural network (CNN) based methods suffer from the demand for large amounts of training data. The bottleneck in prediction accuracy has shifted from data processing limits toward a lack of ground truth samples, usually collected manually by experienced experts. In this work, we provide a metalearning framework for few-shot classification of RS scene. Under the umbrella of meta-learning, we show it is possible to learn much information about a new category from only 1 or 5 samples. The proposed method is based on Prototypical Networks with a pre-trained stage and a learnable similarity metric. The experimental results show that our method outperforms three state-of-the-art few-shot algorithms and one typical CNN-based method, D-CNN, on two challenging datasets: NWPU-RESISC45 and RSD46-WHU.

Convolutional neural network (CNN) based methods have dominated the field of aerial scene classification for the past few years. While achieving remarkable success, CNN-based methods suffer from excessive parameters and notoriously rely on large amounts of training data. In this work, we introduce few-shot learning to the aerial scene classification problem. Few-shot learning aims to learn a model on base-set that can quickly adapt to unseen categories in novel-set, using only a few labeled samples. To this end, we proposed a meta-learning method for few-shot classification of aerial scene images. First, we train a feature extractor on all base categories to learn a representation of inputs. Then in the meta-training stage, the classifier is optimized in the metric space by cosine distance with a learnable scale parameter. At last, in the meta-testing stage, the query sample in the unseen category is predicted by the adapted classifier given a few support samples. We conduct extensive experiments on two challenging datasets: NWPU-RESISC45 and RSD46-WHU. The experimental results show that our method yields state-of-the-art performance. Furthermore, several ablation experiments are conducted to investigate the effects of dataset scale, the impact of different metrics and the number of support shots; the experiment results confirm that our model is specifically effective in few-shot settings.

This paper presents a unified framework for automatically detecting and tracking dim small targets in infrared (IR) image sequence under complex backgrounds. Firstly, the variance weighted information entropy (variance WIE) followed by a region growing technique is introduced to segment the candidate targets in a single-frame IR image after background suppression. Then the pipeline filter is used to verify the real targets. The position and the size of the detected target are then obtained to initialize the tracking algorithm. Secondly, we adopt an improved local binary pattern (LBP) scheme to represent the target texture feature and propose a joint gray-texture histogram method for a more distinctive and effective target representation. Finally, target tracking is accomplished by using the mean shift algorithm. Experimental results indicate that the proposed method can effectively detect the dim small targets under complex backgrounds and has better tracking performance compared with the gray histogram based tracking methods such as the mean shift and the particle filtering.

Recently, Multiple Instance Learning (MIL) technique has been introduced for object tracking\linebreak applications, which has shown its good performance to handle drifting problem. While some instances in positive bags not only contain objects, but also contain the background, it is not reliable to simply assume that each feature of instances in positive bags obeys a single Gaussian distribution. In this paper, a tracker based on online multiple instance boosting has been developed, which employs Gaussian Mixture Model (GMM) and single Gaussian distribution respectively to model features of instances in positive and negative bags. The differences between samples and the model are integrated into the process of updating the parameters for GMM. With the Haar-like features extracted from the bags, a set of weak classifiers are trained to construct a strong classifier, which is used to track the object location at a new frame. And the classifier can be updated online frame by frame. Experimental results have shown that our tracker is more stable and efficient when dealing with the illumination, rotation, pose and appearance changes.

In this paper, we propose a novel object tracking method that can work well in challenging scenarios such as appearance changes, motion blurs, and especially partial occlusions and noise. Our method applies bilateral two-dimensional principal component analysis (Bi-2DPCA) for efficient object modeling and real-time computation requirement. An incremental Bi-2DPCA learning algorithm is proposed for characterizing the appearance changes of newly tracked objects. Also, to account for noise and occlusions, a sparse structure is introduced into our Bi-2DPCA object representation model. With this sparse structure, the appearance of an object can be represented by a linear combination of basis images and an additional noise image. The noise image, which indicates the location of noise and occlusions, can be used to effectively eliminate the influence caused by noise and occlusions and lead to a robust tracker. Instead of the reconstruction error commonly used in eigen-based tracking methods, a more accurate method is adopted for the computation of observation likelihood. The method is based on the energy distribution of coefficient matrix projected by Bi-2DPCA. Experimental results on challenging image sequences demonstrate the effectiveness of the proposed tracking method.

SIFT features have been found to be effective in describing image textures. Because SIFT features have some great characteristics, such as translation invariance, zooming in and out invariance, spin invariance and affine invariance, etc, so the image retrieval precision is satisfactory usually. However, in Content Based Image Retrieval (CBIR), there are so many SIFT feature points extracted from an image and the size of SIFT-based feature vectors can be up to 128 dimensions. So, even though the prevision based on SIFT feature is high, the retrieval speed is low. To relieve this problem, this paper proposes an improved SIFT feature point extraction method. First of all, taking 2-level wavelet transform to the image, then setting its low-frequency sub-band to zero and reconstructing the image by its 6high frequency sub-bands. The SIFT features are then extracted from the reconstructed 'high-frequency images' for retrieval purpose. This method can reduce the number of SIFT feature points by 71.2%. Tested on a tyre tread pattern dataset, the proposed method is found to be able to significantly improve the retrieval speed while the retrieval precision is still better than other existing methods.

Image fusion aims to merge two or more images captured via various sensors of the same scene to construct a more informative image by integrating their details. Generally, such integration is achieved through the manipulation of the representations of the images concerned. Sparse representation plays an important role in the effective description of images, offering a great potential in a variety of image processing tasks, including image fusion. Supported by sparse representation, in this paper, an approach for image fusion by the use of a novel dictionary learning scheme is proposed. The nonlocal self-similarity property of the images is exploited, not only at the stage of learning the underlying description dictionary but during the process of image fusion. In particular, the property of nonlocal self-similarity is combined with the traditional sparse dictionary. This results in an improved learned dictionary, hereafter referred to as the nonlocal sparse K-SVD dictionary (where K-SVD stands for the K times singular value decomposition that is commonly used in the literature), and abbreviated to NL_SK_SVD. The performance of the NL_SK_SVD dictionary is applied for image fusion using simultaneous orthogonal matching pursuit. The proposed approach is evaluated with different types of images, and compared with a number of alternative image fusion techniques. The resultant superior fused images using the present approach demonstrates the efficacy of the NL_SK_SVD dictionary in sparse image representation.

This paper presents a novel algorithm for detecting and tracking dim targets in infrared (IR) image sequence. Firstly the variance weighted information entropy (variance WIE) is introduced for the target detection after background suppression. The position and the size of the detected targets are then obtained to initialize the tracking algorithm. Then we adopt the local binary pattern (LBP) scheme to represent the target texture feature and propose a joint gray-texture histogram method for a more distinctive and effective target representation. Finally, target tracking is accomplished by using the mean shift algorithm. Experimental results indicate that the proposed method can effectively detect the dim targets and achieves much better tracking results compared with the traditional gray histogram based mean shift tracking.

In this paper, the neural network ensemble algorithm is proposed to solve the problem of the mislabeled data in the tri-training process. Firstly, we analyze the advantage of the neural network ensemble, and then introduce it to correct the mislabeled data to improve the quality of the enlarged training set, so the precision and generalization of learns is improved. Experimental results on UCI data sets indicate that the classification performance of the proposed algorithm is 22.87% higher than that of the tri-training algorithm under the four kinds of the unlabeled rates. The proposed algorithm could effectively exploit unlabeled data to enhance the learning performance.

Sparse signal decomposition can get sparse representation of signal. Given that the sparse decomposition has a large number of calculations and is almost impossible to meet the request of real time. A novel multi-swarm co-operative particle swarm optimization (PSO) algorithm to implement matching pursuit was developed, where multi-swarm was adopted to maintain the diversity of population, and the exploration ability of particle swarm optimization was elegantly combined with the exploitation of extremal optimization (EO) to prevent premature convergence. This method could reduce very time-consuming inner product times and improve decomposition accuracy in signal sparse decomposition, thereby, balancing very well search efficiency of time-frequency atoms and computer memory for storing the over-complete dictionary. The results of experiments indicated that the proposed algorithm can effectively speed up the convergence and lead to a preferable solution.

Wavelet transform has the good characteristic of spatial and frequency locality, but it isnpsilat suitable for describing the signals, which have high dimensional singularities. Curvelet is one of new multiscale transform theories, which possess directionality and anisotropy, and it breaks some inherent limitations of wavelet in representing directions of edges in image. So when the curvelet transform is applied in image fusion, the characteristics of original images are taken better and implemented more easily. This paper tries fast discrete curvelet transform (FDCT) for image fusion of SAR (synthetic aperture radar) image and TM (thematic mapper) image. Then, visual result and statistical parameters are used to evaluate the result. The experimental results indicate that the FDCT-based fusion method can provide more detailed spatial information and simultaneously, preserves the richer spectral content than the conventional approach, such as the discrete wavelet transform (DWT) and the intensity-hue-saturation (IHS) transform.

The increasing demands of education, reduction of study cost and study without location limitation have encouraged developers providing different types of learning and educational systems. This paper presents a virtual Chinese hairy brush painting environment which can be used as a digital tool for learning or creating Chinese calligraphy or Chinese painting. The framework comprises two components: brush geometry model and brush dynamic model. Instead of simulating the brush using bristles, we use points to simulate the whole brush bundle, which can drastically decrease the complexity inherent in the conventional bristle-level approach; our brush dynamic model describes the behavior of the real brush's deformation in terms of the interaction of the external and internal forces with the virtual writing paper. A spring network is derived to calculate the physical deflection of brush according to the force exerted on it. A prototype system has been implemented based on this framework, with which high-quality digital painting and calligraphy artwork can be produced.

In this paper, an improved particle swarm optimization algorithm is proposed to train the fuzzy support vector machine (FSVM) for pattern multi-classification. In the improved algorithm, the particles studies not only from itself and the best one but also from the mean value of some other particles. In addition, adaptive mutation was introduced to reduce the rate of premature convergence. The experimental results on MNIST character recognition show that the improved algorithm is feasible and effective for FSVM training.

Multifrequency polarimetric SAR imagery provides a very convenient approach for signal processing and acquisition of radar image. However, the amount of information is scattered in many images, and redundancies exist between different bands and polarizations. Similar to signal-polarimetric SAR image, multifrequency polarimetric SAR image is corrupted with speckle noise at the same time. This paper presents a method of information compression and speckle reduction for multifrequency polarimetric SAR imagery based on kernel principal component analysis (KPCA). KPCA is a nonlinear generalization of linear principal component analysis using kernel trick. The NASA/JPL polarimetric SAR imagery of P, L, and C bands quadpolarizations is used for illustration. Experimental results show that KPCA has better capability in information compression and speckle reduction compared with linear PCA.

An adaptive immune genetic algorithm (AIGA) based on cost minimization technique method for edge detection is proposed. The proposed AIGA recommends the use of adaptive probabilities of crossover, mutation and immune operation, and a geometric annealing schedule in immune operator to realize the twin goals of maintaining diversity in the population and sustaining the fast convergence rate in solving the complex problems such as edge detection. Furthermore, AIGA can effectively exploit some prior knowledge and information of the local edge structure in the edge image to make vaccines, which results in much better local search ability of AIGA than that of the canonical genetic algorithm. Experimental results on gray-scale images show the proposed algorithm perform well in terms of quality of the final edge image, rate of convergence and robustness to noise.

Since training a SVM requires solving a constrained quadratic programming problem which becomes difficult for very large datasets, an improved particle swarm optimization algorithm is proposed as an alternative to current numeric SVM training methods. In the improved algorithm, the particles studies not only from itself and the best one but also from the mean value of some other particles. In addition, adaptive mutation was introduced to reduce the rate of premature convergence. The experimental results show that the improved algorithm is feasible and effective for SVM training.

This paper presents a compound neural network model, i.e., adaptive neurofuzzy network (ANFN), which can be used for identifying the complicated nonlinear system. The proposed ANFN has a simple structure and exploits a hybrid algorithm combining supervised learning and unsupervised learning. In addition, ANFN is capable of overcoming the error of system identification due to the existence of some changing points and improving the accuracy of identification of the whole system. The effectiveness of the model and its algorithm is tested on the identification results of missile attacking area.

The classification of ship targets using the kernel Fisher discriminant analysis is investigated in this paper. The main idea of this method is to find a nonlinear direction by first mapping the data nonlinearly into some feature space and compute Fisher's linear discriminant in input space. Based on the kernel Fisher discriminant, we recognize three types of ships. The satisfactory experimental results are obtained. In addition, we compare this method with other state of the art classification techniques. The experiments show that the kernel Fisher discriminant is superior to the other algorithms.