Hyperspectral Image Classification With Deep Metric Learning and Conditional Random Field

Abstract

To improve the classification performance in the context of hyperspectral image processing, many works have been developed based on two common strategies, namely the spatial-spectral information integration and the utilization of neural networks. However, both strategies typically require more training data than the classical algorithms, aggregating the shortage of labeled samples. In this letter, we propose a novel framework that organically combines the spectrum-based deep metric learning model and the conditional random field algorithm. The deep metric learning model is supervised by the center loss to produce spectrum-based features that gather more tightly in Euclidean space within classes. The conditional random field with Gaussian edge potentials, which is firstly proposed for image segmentation tasks, is introduced to give the pixel-wise classification over the hyperspectral image by utilizing both the geographical distances between pixels and the Euclidean distances between the features produced by the deep metric learning model. The proposed framework is trained by spectral pixels at the deep metric learning stage and utilizes the half handcrafted spatial features at the conditional random field stage. This settlement alleviates the shortage of training data to some extent. Experiments on two real hyperspectral images demonstrate the advantages of the proposed method in terms of both classification accuracy and computation cost.

A versatile deep learning architecture for classification and label-free prediction of hyperspectral images

“…To better reflect the model’s performance, especially in these cases, counts of false positive and negative pixels and the intersection over union (IOU) for each class is provided in Extended Data Table 1 . The overall accuracy (99.48% ± 0.50%), however, is in concordance with state-of-the-art architectures for hyperspectral classification on the Indian Pines dataset 41 , 42 , 55 – 57 . Three of these architectures’ (ResNet, Multi-Path ResNet, and Auxillary Capsule GAN) classification accuracies are shown in Table 1 for comparison with the 17-U UwU-Net demonstrating the highest accuracy.…”

Decision Tree Learning with Spatial Modal Logics

“…Although it proved fruitful in the past, classifying a pixel solely according to its spectral footprint is limited, as it disregards the spatial structure of the image. In recent times there have been many neural network-based applications that can capture spatial, inter-pixel dependencies in some sense [3,14,24,40,42,52]. These approaches leverage intrinsic spatial localities through the engineering of spatial features, and their performances are ultimately better than those of previous attempts; however, as we have already observed, as a general rule functional approaches give up the interpretability and the explainability of the resulting models.…”

Reduced-cost hyperspectral convolutional neural networks

“…28 This can be done using feature extraction or feature selection approaches: the former apply linear or nonlinear transformations to extract specific features from the original data, 29,30 whereas the latter select the most useful individual features (i.e., spectral bands) of the data without transforming it. 31,32 Some approaches based on SVMs using composition of kernels, 33 3-D wavelet filters, 34 3-D Gabor filters, 35 or conditional random fields 36 show improved classification performance, taking into account both spatial and spectral information. Their main drawback, however, is that they require hand-crafted spatial features.…”