Lossy compression is widely used to decrease the size of multichannel remote sensing data. Alongside this positive effect, lossy compression may lead to a negative outcome as making worse image classification. Thus, if possible, lossy compression should be carried out carefully, controlling the quality of compressed images. In this paper, a dependence between classification accuracy of maximum likelihood and neural network classifiers applied to three-channel test and real-life images and quality of compressed images characterized by standard and visual quality metrics is studied. The following is demonstrated. First, a classification accuracy starts to decrease faster when image quality due to compression ratio increasing reaches a distortion visibility threshold. Second, the classes with a wider distribution of features start to “take pixels” from classes with narrower distributions of features. Third, a classification accuracy might depend essentially on the training methodology, i.e., whether features are determined from original data or compressed images. Finally, the drawbacks of pixel-wise classification are shown and some recommendations on how to improve classification accuracy are given.

Lossy compression of remote sensing data has found numerous applications. Several requirements are usually imposed on methods and algorithms to be used. A large compression ratio has to be provided, introduced distortions should not lead to sufficient reduction of classification accuracy, compression has to be realized quickly enough, etc. An additional requirement could be to provide privacy of compressed data. In this paper, we show that these requirements can be easily and effectively realized by compression based on discrete atomic transform (DAT). Three-channel remote sensing (RS) images that are part of multispectral data are used as examples. It is demonstrated that the quality of images compressed by DAT can be varied and controlled by setting maximal absolute deviation. This parameter also strictly relates to more traditional metrics as root mean square error (RMSE) and peak signal-to-noise ratio (PSNR) that can be controlled. It is also shown that there are several variants of DAT having different depths. Their performances are compared from different viewpoints, and the recommendations of transform depth are given. Effects of lossy compression on three-channel image classification using the maximum likelihood (ML) approach are studied. It is shown that the total probability of correct classification remains almost the same for a wide range of distortions introduced by lossy compression, although some variations of correct classification probabilities take place for particular classes depending on peculiarities of feature distributions. Experiments are carried out for multispectral Sentinel images of different complexities.

The subject matter of the article is the methods of automatic clustering of remote sensing data under conditions of a priori uncertainty regarding the number of observed object classes and the statistical characteristics of the signatures of classes. The aim is to develop a method for approximating multimodal empirical distributions of observational data to construct decision rules for pixel-by-pixel statistical classification procedures, as well as to investigate the effectiveness of this method for automatically classifying objects on synthesized and real images. The tasks to be solved are: to develop and implement a procedure for splitting a mixture of basic distributions, while ensuring the following requirements: the absence of a preliminary data analysis stage in order to select optimal initial approximations; a good convergence of the method and the ability to automatically refine the list of classes by combining indistinguishable or poorly distinguishable components of the mixture into a single cluster; to synthesize test images with a specified number of objects and known data distributions for each object; to evaluate the effectiveness of the developed method for automatic classification by the criterion of the probability of correct recognition; to evaluate the results of automatic clustering of real images. The methods used are methods of stochastic simulation, methods of approximation of empirical distributions, statistical methods of recognition, methods of probability theory and mathematical statistics. The following results have been obtained. A method for automatic splitting of a mixture of Gaussian distributions to construct decision thresholds according to the maximal a posteriori probability criterion was proposed. The results of the automatic forming the list of classes and their probabilistic descriptions, as well as the results of the clustering both test images and satellite ones are given. It is shown that the developed method is quite effective and can be used to determine the number of objects’ classes as well as their stochastic characteristics’ mathematical description for pattern recognition tasks and cluster analysis. Conclusions. The scientific novelty of the results obtained is that the proposed approach makes it possible directly during the “unsupervised” training procedure to evaluate the distinguishability of classes and exclude indistinguishable objects from the list of classes.

For recognition of meteorological objects a two-stage algorithm based on the information theory and on the statistical decision theory for a vector space of polarimetric signatures is developed. General description and the results of the algorithm testing performed on modeled radar signals backscattered by specific classes of hydrometeors as well as on the data received in full-scale experiments are given in the paper.

The paper analyses some challenges, that arise at processing of composite images. It classifies the factors influencing tone intensity of images received by optoelectronic surveillance systems. An approach for tone correction of heterogeneous images is substantiated. The paper proves that for investigation of changes in color properties of multi-tone images the hue-saturation-value (HSV) color model is more appropriate.

The subject matter of the article is the methods of morphological spatial filtering of images in pseudo-colors obtained as a result of statistical segmentation of multichannel satellite images. The aim is to study the effectiveness of various methods of post-classification image processing in order to increase the probability of correct recognition for observed objects. The tasks to be solved are: to select a mathematical model describing the training sets of objects’ classes; to implement the procedure of statistical controlled classification by the maximum likelihood method; to evaluate the results of objects’ recognition on the test image by the criterion of the empirical probability of correct recognition; to formalize the procedures of local object-oriented filtering of a segmented image; to investigate the effectiveness of rank filtering as well as weighted median filtering procedures taking into account the results of the classification by k-nearest neighbors in the filter window. The methods used are methods of empirical distributions’ approximation, statistical recognition methods, methods of probability theory and mathematical statistics, methods of local spatial filtering. The following results were obtained. A method for synthesizing a universal mathematical model has been proposed for describing non-Gaussian signal characteristics of objects on multichannel images based on a multi-dimensional variant of Johnson SB distribution; this model was used for statistical pixel-by-pixel classification of the original satellite image. Algorithms for local post-classification processing in the neighborhood of the selected segments boundaries have been implemented. The analysis of the developed algorithms’ effectiveness based on estimates of classes’ correct recognition probability is performed. Conclusions. The scientific novelty of the results obtained is as follows: combined approaches to the pattern recognition procedures have been further developed – it has been shown that the use of methods of local object-oriented filtering of segmented images allows to reduce the number of point errors for element-wise classification of spatial objects.