Multi-class remote sensing object recognition based on discriminative sparse representation

Abstract

The automatic recognition of multi-class objects with various backgrounds is a big challenge in the field of remote sensing (RS) image analysis. In this paper, we propose a novel recognition framework for multi-class RS objects based on the discriminative sparse representation. In this framework, the recognition problem is implemented in two stages. In the first, or discriminative dictionary learning stage, considering the characterization of remote sensing objects, the scale-invariant feature transform descriptor is first combined with an improved bag-of-words model for multi-class objects feature extraction and representation. Then, information about each class of training samples is fused into the dictionary learning process; by using the K-singular value decomposition algorithm, a discriminative dictionary can be learned for sparse coding. In the second, or recognition, stage, to improve the computational efficiency, the phase spectrum of a quaternion Fourier transform model is applied to the test image to predict a small set of object candidate locations. Then, a multi-scale sliding window mechanism is utilized to scan the image over those candidate locations to obtain the object candidates (or objects of interest). Subsequently, the sparse coding coefficients of these candidates under the discriminative dictionary are mapped to the discriminative vectors that have a good ability to distinguish different classes of objects. Finally, multi-class object recognition can be accomplished by analyzing these vectors. The experimental results show that the proposed work outperforms a number of state-of-the-art methods for multi-class remote sensing object recognition.

Vessel detection and classification from spaceborne optical images: A literature survey

“…Processing time is especially significant for real-time vessel detection applications, when results need to be delivered quickly. Quite some number of authors listed the time consumption of their algorithms (Bi et al, 2010, Ding et al, 2012, Dong et al, 2013, Guang et al, 2011, Ji-yang et al, 2016, Li et al, 2016, Lin et al, 2012, Tang et al, 2015, Wang et al, 2016, Xiaoyang et al, 2016, Xu and Liu, 2016, Yang et al, 2017, Yang et al, 2014, Yao et al, 2016). The time spent for calculation is strongly dependent on the computer hardware and the results are thus not always comparable.…”

“…fall under the label ‘Target detection’. Several authors (Han et al, 2014, Li and Itti, 2011, Lin et al, 2016, Wang et al, 2016, Yokoya and Iwasaki, 2015, Zhang et al, 2016) provide no explicit description of the vessel detection purpose but a general one for target detection. In the case of such occurrence, the analysis in this review takes into account only vessel detection.…”

“…Liu et al2016Google EarthGoogle Earth−−n/aStatistical method: rotated bounding box spaceDiscrimination: binary linear modellingyesVessel detectionZ. Liu et al2016Google Earthn/a−−n/an/aDiscrimination: attribute-based modelyesVessel category recognitionOliveau and Sahbi2016Google Earth, Gaofen-1n/a−1 (GE), (GF-1)n/aThreshold-based method: Otsu segmentation, morphological operationsDiscrimination: based on geometric characteristicsyesVessel detectionShuai, Sun, Shi, Chen2016Google Earthn/a−−n/aComputer vision method: scale invariant feature algorithmDiscrimination: maximum match number with libraryyesVessel detectionShuai, Sun, Wu et al2016WorldView-2, QuickBird-2, GeoEye-1, RapidEye, Formosat-2, Sentinel-2PAN, B, G, R, NIR0.5 (WV), 0.6 (QB), 0.4 (GE), 6.5 (RE), 2 (FS)2 (WV), 2.4 (QB), 1.6 (GE), 5 (RE), 8 (FS), 10 (S2)0–15Shape and texture: Minimum Noise Fraction algorithm, object-based image analysisClassification: threshold-rule set classificationyesVessel detection for immigrant search and rescueTopputo et al2016Google Earthn/a−−n/aComputer vision method: scale invariant feature transform descriptor with improved bag-of-words modelDiscrimination: phase spectrum of quaternion Fourier transformyesTarget detectionX. Wang et al2016Orbview-3PAN, B, G, R14n/aAnomaly detection method: Probability density function, vessel distribution by the densityDiscrimination: structural continuity descriptor (based on width to length ratio)yesVessel detection on open seaXiaoyang et al2016Google EarthPAN−−n/aSalient-based method: hypercomplex frequency domain and phase quaternion Fourier transformDiscrimination: radon transform and histogram of oriented gradientsyesVessel de...…”

Automatic Ship Detection in Optical Remote Sensing Images Based on Anomaly Detection and SPP-PCANet

“…In recent years, the extraction approaches of ship candidate areas have been proposed, for example, bayesian decision (BD) [23], compressed domain [24], and convolutional neural network [16,25]. In addition, the sparse feature method based on multi-layer sparse coding [26][27][28][29][30] was used to segment the saliency map [21,31,32] to obtain candidate regions. An effective multi scale CFAR detector for the gamma distribution clutter model [33] was also designed to detect candidate targets in the sea.…”

Detecting Saliency in Infrared Images via Multiscale Local Sparse Representation and Local Contrast Measure

“…In recent years, the sparse representation (SR) theory has been developed in the field of signal processing and analysis [11]. Image sparse representation is the linear representation of image signal by a few atoms in the learned overcomplete dictionary.…”