遥感探测到的小目标信号一般是弱信号，利用传统的高光谱异常变化检测方法直接抑制背景来突出异常变化目标，往往导致小目标弱信号同时被抑制，造成目标探测率低、虚警率高。基于独立成分分析方法，研究了弱信号小目标的高光谱变化检测模型，该模型首先通过投影寻踪将异常变化影像投影到独立成分，突出异常变化目标，然后再抑制背景，从而达到异常变化目标和背景的有效分离。该模型可以有效降低虚警率，提高探测率。利用模拟数据和真实数据进行了精度验证，结果表明，利用模拟数据得到的探测精度为99%，利用真实数据得到的检测精度为86%，与传统异常变化检测算法相比，精度最高提高了9%。本文研究方法适用于弱信号小目标的高光谱异常变化检测。

Small target signals detected through remote sensing are typically weak signals. The traditional hyperspectral anomaly change detection method directly suppresses the background. However, it frequently causes small targets to be suppressed simultaneously, thereby resulting in a low target detection rate and a high false alarm rate. In this study, a hyperspectral anomaly change detection model based on independent component analysis is used. The proposed model is projected on an independent component, which first highlights the anomaly changes and then suppresses the background to achieve the effective separation of anomaly changes and background. This model can effectively reduce the false alarm rate and improve the detection rate. The accuracy is verified by simulation and real data. Results show that the detection accuracy is 99% using simulated data, and the detection accuracy is 86% using real data. The accuracy is increased by 9% in comparison with the traditional anomaly change detection algorithm. The proposed hyperspectral anomaly change detection method is suitable for processing weak targets.
Remote sensing image change detection is the process of quantitatively analyzing the surface changes in remote sensing images that are not obtained in the same surface area simultaneously. However, change detection frequently fails to highlight the change in interest given the differences in atmospheric environment and radiation difference caused by various sensors. We aim to find the small changes that are rare and different from the overall background trend.
Traditional hyperspectral anomaly detection methods are generally used to highlight abnormal changes by directly suppressing the background. However, the three methods mentioned above cannot effectively eliminate the radiation differences in the case of complex objects and cannot guarantee the consistency of the background. The background is difficult to suppress, and the anomaly changes cannot be highlighted.
Abnormal change detection method based on Independent Component Analysis (ICA) through the abnormal changes in RX anomaly detection method of pixels selects the anomalies with strong changes in abnormal pixels for initial projection direction and to all pixels to initialize the orthogonal projection of projection direction and abnormal pixel labeling for a second projection direction until the number of iterations to achieve independent component.
The visual discrimination results after joining LCRA anomaly change detection results effectively restrain the false alarm rates and highlight the anomaly change targets, and abnormal change detection obtains accurate results. The ICA results show that the accuracy of anomaly change detection is superior to other methods.
The result shows the quantitative evaluation result of abnormal change detection. ICA realizes the highest accuracy. The ICA anomaly change detection method achieves a low false alarm rate and favorable detection effect, and this method can obtain the highest accuracy with or without the LCRA matching strategy. Considering the evident geometric matching error of real data, the accuracies of all methods improve by approximately 0.2 after the LCRA matching strategy is adopted.
The result shows the corresponding accuracy of the ICA anomaly detection method for analyzing the number of different independent components. Moreover, the result accuracy of this method is the highest when the number of independent components is moderate and between 7 and 14, thereby reflecting its robustness. The accuracy of the proposed method is better than other methods when the parameter size is reasonable.
This study presents an ICA model, which determines the abnormal changes in projection on an independent component and the prominent changes in the target and restrains the background to achieve the effective separation of target and background, effectively improve the detection rate, and achieve low false alarm rates. The conclusions are summarized as follows: (1)The proposed method that uses simulated data achieves abnormal change detection accuracy of 99% and real data detection accuracy of 86% in comparison with the traditional abnormal change detection algorithm, which accuracy reaches 9%; (2) Aiming at the abnormal change target of a subpixel level, the proposed method has a 1% improvement in accuracy in comparison with traditional abnormal change detection.
(3) The proposed method has only one parameter, and the selection of parameters slightly impact accuracy, which has strong robustness.