高光谱图像分类是图像解译任务的重要技术之一,已经在遥感观测、智慧医疗等诸多领域得到广泛的应用。本质上,高光谱图像分类由特征提取与基于分类器的标签预测这两阶段操作组成。现有分类方法在特征提取时,大多不考虑分类器的影响,会导致提取的特征与所用分类器之间的兼容性较差,难免出现预测结果差的情况。针对此问题,本文提出具有分类器机制的高光谱图像特征提取方法,保证特征提取与分类器之间的兼容性,使特征能更易于被分类器准确计算,改善分类预测结果。本文给出了两种具有分类器机制的高光谱图像特征提取模型的形式: 1)以稀疏表示和支持向量机为例,将支持向量机特性集成到稀疏表示形式中,建立了能够与支持向量机分类器相兼容的SRS特征提取模型；2)以深度自编码网络与softmax函数为例,将softmax分类器特性嵌入到深度自编码网络中,构建能与softmax分类器相兼容的DAES特征提取模型。为获得SRS和DAES模型的解,本文还给出了对应的求解策略与优化过程。在遥感高光谱图像和医学高光谱图像数据上开展实验验证,结果表明,本文SRS和DAES算法具有明显的有效性和优越性,在高光谱图像分类指标OA(overall accuracy)、AA(average accuracy)、Kappa上分别提升约5.03%、5.13%、7.30%。

Objective: As an important technique in the image interpretation task, hyperspectral image (HSI) classification has been widely used in many fields such as remote sensing observation and intelligent medical service. Actually, the HSI classification can be seen as consisting of feature extraction and classifiers based label prediction. Most of current classification approaches have no consideration of the influence of classifiers on feature extraction, which may cause the bad compatibility between the extracted features and the used classifier and have poor prediction results. Method: In order to remedy such deficiency, this paper presents a novel kind of HSI feature extraction methods embedded by the classifier mechanism, which can ensure the compatibility between feature extraction and the used classifier, so that features can be more easily calculated by classifier accurately, and classification prediction results can be thus improved. Two specific forms are given in this paper.1) The SRS feature extraction model compatible with support vector machine (SVM) classifier is built, which embeds the SVM property into the sparse representation (SR). 2) The DAES feature extraction model compatible with softmax classifier is constructed, which integrates the softmax function into the deep autoencoder (DAE) network. We also provide the optimization strategy to obtain the solutions of the SRS and DAES models. Results: The proposed SRS and DAES models are experimentally evaluated on the remote sensing HSI data and medical HSI data, and the experiments consists of parameter analysis, algorithm comparison, ablation study, and convergence analysis. According to the parameter analysis, we validate that the values of important parameters have obvious impact on the performance of our methods, and successfully select the best values of these parameters. As suggested by the algorithm comparison, the proposed methods achieve the better classification performance compared with some state-of-the-art approaches, which have obvious effectiveness and superiority, which are averagely higher 5.03%, 5.13%, 7.30% at OA (overall accuracy), AA(average accuracy), and Kappa indexes in the HSI classification task. The ablation study is conducted to demonstrate the effectiveness of the compatibility between feature extraction and the bedded classifiers. The convergence analysis indicates that the designed optimization solution strategy can meet the application requirements of reliability and rapidity. Conclusion: As discussed in the above-description, it can be concluded that the proposed SRS and DAES methods realize the good compatibility between feature extraction and classifiers, so that the extracted features can be better calculated by classifiers, and the more competitive classification performance can be achieved.