Classification is a common objective when analyzing hyperspectral images, where each pixel is assigned to a predefined label. Deep learning-based algorithms have been introduced in the remote-sensing community successfully in the past decade and have achieved significant performance improvements compared with conventional models. However, research on the extraction of sequential features utilizing a single image, instead of multi-temporal images still needs to be further investigated. In this paper, a novel strategy for constructing sequential features from a single image in long short-term memory (LSTM) is proposed. Two pixel-wise-based similarity measurements, including pixel-matching (PM) and block-matching (BM), are employed for the selection of sequence candidates from the whole image. Then, the sequential structure of a given pixel can be constructed as the input of LSTM by utilizing the first several matching pixels with high similarities. The resulting PM-based LSTM and BM-based LSTM are appealing, as all pixels in the whole image are taken into consideration when calculating the similarity. In addition, BM-based LSTM also utilizes local spectral-spatial information that has already shown its effectiveness in hyperspectral image classification. Two common distance measures, Euclidean distance and spectral angle mapping, are also investigated in this paper. Experiments with two benchmark hyperspectral images demonstrate that the proposed methods achieve marked improvements in classification performance relative to the other state-of-the-art methods considered. For instance, the highest overall accuracy achieved on the Pavia University image is 96.20% (using both BM-based LSTM and spectral angle mapping), which is an improvement compared with 84.45% overall accuracy generated by 1D convolutional neural networks.

Extracting endmembers from remotely-sensed images of vegetated areas can present difficulties. In this research, we applied a recently-developed endmember-extraction algorithm based on Support Vector Machines to the problem of semi-autonomous estimation of vegetation endmembers from a hyperspectral image. This algorithm, referred to as Support Vector Machine-Based Endmember Extraction (SVM-BEE), accurately and rapidly yields a computed representation of hyperspectral data that can accommodate multiple distributions. The number of distributions is identified without prior knowledge, based upon this representation. Prior work established that SVM-BEE is robustly noise-tolerant and can semi-automatically estimate endmembers; synthetic data and a geologic scene were previously analyzed. Here we compared the efficacies of SVM-BEE, N-FINDR, and SMACC algorithms in extracting endmembers from a real, predominantly-agricultural scene. SVM-BEE estimated vegetation and other endmembers for all classes in the image, which N-FINDR and SMACC failed to do. SVM-BEE was consistent in the endmembers that it estimated across replicate trials. Spectral angle mapper (SAM) classifications based on SVM-BEE-estimated endmembers were significantly more accurate compared with those based on N-FINDR- and (in general) SMACC-endmembers. Linear spectral unmixing accrued overall accuracies similar to those of SAM.

Self-Organizing Map-based Applications in Remote Sensing | InTechOpen, Published on: 2010-04-01. Authors: Anthony Filippi, Iliyana Dobreva, Andrew G. Klein, et

Riparian forests are critical for carbon storage, biodiversity, and river water quality. There has been an increasing use of very-high-spatial-resolution (VHR) unmanned aircraft systems (UAS)-based remote sensing for riparian forest mapping. However, for improved riparian forest/zone monitoring, restoration, and management, an enhanced understanding of the accuracy of different classification methods for mapping riparian forests and other land covers at high thematic resolution is necessary. Research that compares classification efficacies of endmember- and object-based methods applied to VHR (e.g., UAS) images is limited. Using the Sequential Maximum Angle Convex Cone (SMACC) endmember extraction algorithm (EEA) jointly with the Spectral Angle Mapper (SAM) classifier, and a separate multiresolution segmentation/object-based classification method, we map riparian forests/land covers and compare the classification accuracies accrued via the application of these two approaches to narrow-band, VHR UAS orthoimages collected over two river reaches/riparian areas in Austria. We assess the effect of pixel size on classification accuracy, with 7 and 20 cm pixels, and evaluate performance across multiple dates. Our findings show that the object-based classification accuracies are markedly higher than those of the endmember-based approach, where the former generally have overall accuracies of >85%. Poor endmember-based classification accuracies are likely due to the very small pixel sizes, as well as the large number of classes, and the relatively small number of bands used. Object-based classification in this context provides for effective riparian forest/zone monitoring and management.