Time-series of satellite images may reveal important data about changes in environmental conditions and natural or urban landscape structures that are of potential interest to citizens, historians, or policymakers. We applied a fast method of image analysis using Self Organized Maps (SOM) and, more specifically, the quantization error (QE), for the visualization of critical changes in satellite images of Las Vegas, generated across the years 1984-2008, a period of major restructuration of the urban landscape. As shown in our previous work, the QE from the SOM output is a reliable measure of variability in local image contents. In the present work, we use statistical trend analysis to show how the QE from SOM run on specific geographic regions of interest extracted from satellite images can be exploited to detect both the magnitude and the direction of structural change across time at a glance. Significantly correlated demographic data for the same reference time period are highlighted. The approach is fast and reliable, and can be implemented for the rapid detection of potentially critical changes in time series of large bodies of image data.
The analysis of time series of images by computational methods or algorithms represents a complex challenge in science and society. The detection and characterization of critical changes in public spaces of the natural or the built environment reflected by changes in image time series such as photographs or remotely sensed image data may be of considerable importance for risk mitigation policies and public awareness. This places a premium on fast automatic techniques for discriminating between changed and unchanged contents in large image time series, and computational methods of change detection in image data including remotely sensed data, exploiting different types of transforms and algorithms, have been developed to meet this challenge. Existing methods have been reviewed previously in excellent papers by [1] and [2]. Known computations include Otsu's algorithm [3], Kapur's algorithm [4], and various other procedures such as pixel-based change detection, image differencing, automated thresholding, image rationing, regression analysis on image data, the least-square method for change detection, change vector analysis, median filtering, background filtering, and fuzzy logic algorithms [5][6][7][8][9][10][11]. The scope of any of these methods is limited by the specific goal pursued; for detailed reviews, see [1,2].
In general terms, change detection consists of identifying differences in the state of an object or phenomenon by observing it at different times and implies being able to quantify change(s) due to the effect of time on that given object or phenomenon. In image change detection this involves being able to reveal critical changes through analysis of discrete data sets drawn from image time series. One of the major applications of change detection concerns remotely sensed data obtained from Earth-orbiting satellites. These provide image time series through repetitive coverage at short intervals with consistent image quality, as shown previously in [1]. In this study here, we use a fast, unsupervised change detection technique based on the functional principles of self-organizing maps (SOM), first introduced by Kohonen [12]. Extracts from satellite images representing specific geographic regions of interest of Las Vegas County were used as input to SOM. After preprocessing to ensure equivalence in scale and contrast intensity of the extracted images within a time series, the image input is exploited directly without additional or intermediate procedures of analysis, bridging a gap between classic machine learning and traditional methods of geographic image analysis [13,14]. Our previous work [15] had shown that a specific output variable of the SOM, the quantization error (QE), can be exploited as a diagnostic indicator for the presence of potentially critical local changes in medical image contents. In the present work, we provide proof-of concept simulations showing that the QE in the SOM output is significantly sensitive to spatial extent and intensity of local contrast differences in images. To control for differences in intensity across images of a given time series, a transform is applied before running SOM on each image of the time series. Thereafter, the QE from the SOM output provides a statistically reliable indicator of changes in the spatial extent of contrast regions across image contents, as will be shown. We then use the QE output from SOM on adequately preprocessed extracts from satellite images of Las Vegas County generated across the years 1984-2008. The image extracts correspond to two distinct geographic regions of interest (ROI) here: Las Vegas City Center and Lake Mead and its close surroundings in the Nevada Desert. The reference time period chosen for this study here is of particular interest because of 1) major structural changes in the urban landscape of Las Vegas City during that period, and 2) the gradual dwindling of Lake Mead’s water levels due to the effects of global climate change. We use statistical trend analysis to prove that the QE from the SOM on the different image ROI reliably reflects these critical changes across the years. Using Pearson’s correlation analysis, we show that the QE output is significantly correlated with the most relevant demographic data for the same reference time period.
This section is divided into three parts. In the first, results from a set of proof of concept studies are shown, with SOM run on monochromatic images with increasing spatial extent of contrast at constant intensity (first series of six images), and on images with increasing contrast intensity at constant spatial extent of contrast (second series of six images). This part is to show the statistically significant sensitivity of the QE to the spatial extent and the intensity of contrast in image series. In the first and the third parts, QE results from SOM run on the time series of images for the two geographic ROI are shown and discussed in the light of their statistic
This content is AI-processed based on open access ArXiv data.
