Digital signal and image processing and analysis

· Telemetric information processing

- Methods and algorithms for intelligent processing of telemetric information coming from onboard objects via radio channel based on modeling their behavior. The intellectual basis of the developed algorithms are methods for simulation of the state and dynamics of telemetric objects, neural network methods of analysis and prediction and methods of hierarchical neural network classification. The scientific significance of the results consists in the development of a two-level model of heterogeneous ensembles neural networks (ENN) using elements of the evolutionary strategy in teaching; a two-level model of heterogeneous ANNs with different time scales of the input data.

- The algorithm for incremental learning of ENN identification for the synthesis of a multilayer perceptron, which avoids the heuristic process of choosing the sizes of hidden layers in a two-level ENN training model with non-stationary input data and increases the overall classification accuracy due to a step-by-step concentration on poorly identifiable data. The algorithm for the formation of training samples is proposed in order to improve the accuracy of the analysis by reducing the redundancy of the processed data. The algorithm includes: removal of time series that take constant values ​​throughout; removal of outliers; handling missing values; performing data scaling; construction of an auto-associative neural network designed for nonlinear reduction of the input data dimension; division of the processed dataset into training, test and validation samples.

- Technique of testing ENN for a comparative analysis of various approaches to constructing ENN for predicting the state of technical objects based on telemetry data. Formation of training examples is carried out according to the sliding window principle. When preparing the general set of telemetric information, resampling (conversion of the initial data into a form with a fixed sampling time) and scaling (bringing the data into the range [‑1, 1]) are performed. The network outputs are also scalable. The input set is divided in a 9:1 ratio into a general training and final test set. The general training sample was randomly divided into validation (15%), test (15%), and training (70%) samples. The mean square of the error and the mean absolute error are used to assess the quality of trained ENN and compare different ENN architectures.

Basing on these algorithms, the prototype (within the framework of the STP US "Kosmos-NT") and the experimental sample of the neural network system for monitoring the state and behavior of spacecraft subsystems based on telemetry data for the ground command and measurement complex (within the framework of the STP US "Monitoring-SG") were developed.

[IN101, IK403, IK704, F11OB-071, 6MSG/13-222, Grant-NASB, F15M-059]

· Digital image processing

- The method for identifying topology objects based on a special class of segmentation algorithms using a discrete orthogonal transformation in the basis of two-dimensional functions, a modified Hough transform and a cluster approach using new distance functions and pixel space discretization for images of integrated circuit’s topological layers. The method is characterized by resistance to changes in shooting conditions of various image frames, which ensures the reliability of identification up to 95% when restoring the topology and identifying defects.

- Neural network models in the form of a multilayer perceptron and neocognitron for identification and classification of objects of integrated circuit topological layers and algorithms for their training using the Lipschitz criterion that provide guaranteed convergence of training and recognition stability at the level of 94.55%.

- Algorithms for stitching raster frames of topological images of integrated circuit layers based on special heuristics that ensure the correct alignment of overlapping areas without reference marks and restoration of the topology with an accuracy of one pixel.

- Technology for processing images of topological layers of integrated circuits for optical control of topological layers based on original approaches to the formation of defect descriptors, restoration of topology and modification of logic circuits by decomposition of their functional and structural descriptions, the use of which allows to reduce labor costs in the design and manufacture of electronic products.

- Software architecture of computer systems for processing topological layers of integrated circuits and schemes for parallelizing processing algorithms based on the multi-agent approach, providing additional acceleration of image processing by 8% due to dynamic optimization of schedules.

The results were implemented in the STC "Belmikrosystems" of JSC "Integral" in solving the problems of reverse engineering of VLSI, SPRUE “KBTEM-OMO” of SSPA TM "Planar" in the development of algorithms for the formation and control of topological structures on photomasks and semiconductor wafers in the production of VLSI, in JSC "Brest Radio Engineering Plant" for quality control of PCB assembly; in research laboratory 4.6 "Integrated micro- and nanosystems" of BSUIR in the manufacture of systems on a wafer based on basic matrix crystals using the technology of templateless photolithography with the effect of reducing labor costs and import substitution.

- Models, methods and algorithms for processing and recognizing objects of structurally complex images. Algorithms for increasing the information content of semiconductor microcircuit layers color images based on color segmentation and highlighting objects of interest on several frame sets of one VLSI layer with further stitching into a new layer image. The algorithms are distinguished by the use of both tracks and contact windows and the VLSI layer as objects of interest. The area of ​​unidentified image areas is reduced during the operation of the algorithm by combining the segments corresponding to the tracks, contact windows and the layer substrate of several image sets of the processed layer.

- Methods and algorithms for identifying informative features based on geometric invariants, vectorization algorithms based on the modified Hough transform, neural network algorithms for image clustering and classification of objects characterized by a fuzzy description of the shape, size and location.

- The noisy image search algorithm based on hit-or-miss transformation and featuring soft erosion and a semantic filter. The analysis of the algorithm's operation showed high search results both in an image without noise and in a noisy image with and without preliminary filtering. The algorithm is designed to prepare data for experimental studies of the effectiveness of combined technologies for neural network classification of objects on topological layers of integrated circuits.

- Algorithm for color segmentation of semiconductor microcircuits layer images, based on the formation of clusters by the operator based on the results of k-means clustering in RGB space. A feature of the algorithm is the possibility of its problem-oriented adjustment for preprocessing and post-processing using semantic and morphological filters. The algorithm is an integral part of the combined technology for processing images of integrated circuit topological layers.

- The algorithm for image preprocessing with respect to the optical characteristics of scanning systems, which realizes brightness equalization in one frame for the entire layer based on the HSV and RGB color model. The algorithm makes it possible to eliminate brightness inhomogeneity in the image of a microcircuit layer. This is necessary for the subsequent correct microcircuit layer image processing: stitching frames and topology objects identification.

- The method for increasing the accuracy of PCB image segmentation based on expert classification of clustering results by k-means in RGB space. The method improves the segmentation accuracy at the boundaries of topology elements and improves the quality of the analysis of the printed circuit board image.

- Methodology for determining the class of cluster procedures, which includes: the choice of a fuzzy clustering algorithm based on a priori assumptions about the type of cluster partition; determination of the quality indicators of the partition; determining a method for initializing the initial values ​​of the partition; evaluation of a number of partitions using the selected indicator for various parameters of the selected algorithm and highlighting the best parameters for subsequent image processing. This methodology makes it possible to determine the fuzzy clustering algorithm based on a specific thematic problem and in accordance with the characteristics of the data being processed.

- The model of image processing in the form of sequential synthesis of high-resolution color images based on high-resolution panchromatic imagery and low-resolution multispectral images, calculating a basic set of features using a direct relational heuristic algorithm of feasible clustering, identification and classification based on a weighted ensemble of multilayer neural networks with one a hidden layer with a non-linear activation function hyperbolic tangent and an output layer with a soft maximum activation function. Such a model makes it possible to increase the accuracy in comparison with a single neural network and to work with a compressed indicative description of areal image objects. The experimental software is implemented in the Python language and is intended for organizing the experimental verification of the proposed synthesis technique.

- Experimental software for detecting areas of the earth's surface affected by natural disasters is based on the use of SegNet and U-Net neural network models, which made it possible to achieve an accuracy of 90-93% (for U-Net). This software is implemented using the Keras library (the Tensorflow library was used as a computational library).

A. Dudkin. Image processing in the design and production of integrated circuits / A. Dudkin, R. Sadychaŭ. – Minsk: UIIP NAS of Belarus, 2008. – 270 p.

[IN101, IK403, IK704, F11OB-071, 6MSG/13-222, F18V-005, F16M-023]

· Hyperspectral data compression

- Algorithm of context modeling based on adaptive Huffman codes and the model for compressing hyperspectral data of Earth remote sensing based on it. The algorithm allows to increase the data compression ratio by 6.2% compared to the context-adaptive QM codec. The model of compressing hyperspectral data of Earth remote sensing using the context modeling algorithm can be applied both to the spectral channels of the classical hypercube and to Fourier interferograms. The practical significance of the developed modeling algorithm is its software implementation and the possibility of its use in the development of models (schemes) of compression of various types of data. The developed compression model responsive to the context modeling algorithm can serve as the basis for the development of hardware and software tools for compressing hyperspectral data, which determines its economic significance.