Object recognition and scene analysis

The algorithm for modeling and motion analysis of the parameterized objects, the modified algorithm for calculating the optical flow, the algorithm for estimating the parameters of the flow of point objects at large interframe displacements, the algorithm for modeling background pixel values ​​on a dynamic image, the algorithm for detecting and tracking the movement of objects specified parametrically (based on threshold estimates of the differences frames in a video sequence and uses correlation filters for complete restoration of the shape (classification) of moving objects, object parameters (initial data) determine the construction of correlation masks, and correlation peaks allow classifying moving objects). The algorithms are implemented as a software system for processing, analyzing and classifying dynamic scenes. The system is implemented in a component model with an open architecture. This made it possible to use common tools for modeling dynamic scenes with parameterized objects and motion paths – Autodesk 3ds Max, AutoDesk Maya and the like. The possibilities of synthesizing dynamic scenes can be extended using the OpenGL, GDI+ libraries in the C ++ Builder, Visual C ++, Visual C # development environments. The algorithms are focused on use in intelligent video surveillance systems that are used for security and control. Potential consumers of such systems are government services, city departments and enterprises. (INFOTECH 60).

[INТ60]

· Identification and recognition of objects in images

Methods and algorithms for processing and identifying images for building maps of agricultural crops, differing in the use of spectral models at the stage of preliminary image processing and the extraction of informative features, and models of artificial neural networks for color separation and image analysis based on fuzzy logic. On their basis, the computer technology that significantly increases the reliability of recognizing vegetation areas with specified color characteristics and detecting changes in the observed area at various stages of plant disease has been created. The proposed technology can be used as the basis for systems for environmental monitoring, change control and mapping of vegetation in agriculture and forestry.

The technique for recognizing objects in aerial photographs of agricultural vegetation for monitoring diseases. This technique includes the selection of the most informative features and properties of classified objects based on the interval-valued representation of the initial data about objects, the construction of object descriptors based on fuzzy logic, clustering by one of the fuzzy and possible clustering algorithms and the construction of a fuzzy inference system of the Mamdani type, the use of an inference system for classification of pixels extracted from the input images. In this case, the actual recognition of objects in aerial photographs of agricultural vegetation is performed in a fuzzy inference system. When monitoring vegetation diseases via this technique, a sequence of inference systems that corresponds to a change in the state of vegetation is generated.

Algorithms for identifying diseased areas of agricultural vegetation. They are based on the analysis of the histograms of the original images presented in RGB and HSV form. Using color characteristics of images of agricultural fields made it possible to highlight areas of the fields affected by the disease when the first visible signs of the disease appeared.

The method for combining informative features of multispectral images to assess the state of agricultural vegetation. The method is based on the combined use of data in the visible range and a number of vegetation indices calculated from images in the visible and infrared regions of the spectrum, as well as color and texture characteristics. The use of the proposed method makes it possible to form various thematic maps for the tasks of precision farming.

The algorithm for recognizing the state of agricultural vegetation according to aerial photography data based on a combination of two convolutional neural networks used as a classifier for individual image areas ("soil"–"vegetation"–"healthy vegetation"–"affected vegetation"). Further the results of the work of classifiers are combined, which gives three classes of images. The use of the proposed recognition algorithm made it possible to increase the accuracy of the experimental recognition system developed in the systems identification laboratory of the UIIP NAS of Belarus.

The experimental recognition system. Testing of the system showed that the NDVI index based on the near infrared spectral range is the most informative, and also that the use of a neural network model makes it possible to form the contours of recognized objects more accurately. As the result’s development, the following directions have been determined: the use of additional image sensors; the use of additional phenotypic spectral indices of vegetation; application of deep neural networks of semantic segmentation; learning rate optimization for neural networks.

The technique for recognizing objects in aerial photographs of agricultural vegetation for monitoring diseases. This technique includes the selection of the most informative features and properties of classified objects based on the interval-valued representation of the initial data about objects, the construction of object descriptors based on fuzzy logic, clustering by one of the fuzzy and possible clustering algorithms and the construction of a fuzzy inference system of the Mamdani type, the use of an inference system for classification of pixels extracted from the input images. Along with this, the actual recognition of objects in aerial photographs of agricultural vegetation is performed in a fuzzy inference system. When monitoring vegetation diseases, a sequence of inference systems that corresponds to a change in the state of vegetation is generated using this technique.

Experimental software for recognizing objects on aerial photographs of agricultural vegetation, which consists of modules for user interaction, preprocessing, post-processing, recognition, processing and MongoDB database. The software is implemented in Python. The built-in facilities of the OpenCV library are used for visualization. The developed software is designed to organize experimental verification of the proposed object recognition algorithms, and architectural solutions allow to connect image preprocessing, analysis and post-processing modules without changing the program code, which makes it possible to flexibly customize the order of actions performed by the modules. Experimental software for recognizing objects on aerial photographs of agricultural vegetation has been developed using convolutional neural networks based on four-class neural network models SegNet and U-Net. The presented software made it possible to perform segmentation of aerial photographs with an average accuracy of 92-93%. The results of the work can find practical use in systems for processing data of Earth remote sensing for solving problems of precision farming, sustainable forest management and vegetation state monitoring.

[F16MS-012, F18V-005, IK704, F18PLShG-008]

· Topology control on PCB images

Topological image recognition technology based on the topology image decomposition into elementary fragments and their further analysis. The technology includes:

- building the topology model that describes conductors, contact pads and printed circuit board defects with a set of elementary fragments, their characteristics and connections;

- the method and the algorithm for identifying informative features based on the decomposition of the printed circuit board topology into fragments;

- the algorithm for the formation of a fuzzy description of defects in the topology of printed circuit boards based on the calculation of its informative parameters and their representation in the form of an interval-valued fuzzy set;

- the method for increasing the accuracy of PCB images segmentation based on an expert classification of clustering results by k-means in RGB space;

- the neural network method for searching and classifying defects in the topology of printed circuit boards based on the use of a fuzzy objects description and the proposed algorithm for its formation;

- the algorithm for training an ensemble of neural networks to classify topology defects on images of printed circuit boards;

- the hybrid algorithm for controlling the topology of a printed circuit board based on the analysis of the shape and coordinates of its elementary fragments – contact pads and rectangular fragments of tracks, and checking the correspondence of their connections to the corresponding connections of the reference printed circuit board;

- the algorithm for seeking macro defects in the topology of a printed circuit board based on the combination of local sets of topology defects found at the preliminary stages of its control by the method of morphological closure;

- the algorithm for searching for macro defects of the PCB substrate based on the analysis of the color and brightness characteristics of the substrate image.

The proposed technology is implemented in the form of an experimental software system for searching and classifying defects in the topology of printed circuit boards.

[F16-070, F19MC-032]

· VLSI semiconductor wafer topology control

Methods and algorithms for improving images of semiconductor wafers; formation of defect descriptors; qualitative and quantitative descriptions of topology defects used in the technological process of integrated circuits manufacturing. On the basis of this, The software for classifying topology defects detected revealed during automatic control of topological layers of semiconductor wafers has been developed. The mentioned topological layers implement the functions of managing the defect database, classifying and visually viewing defects on the digital model of the wafer, viewing defect lists and statistical analysis of defects.

[IK403, F19MC-032, F16-070, IIT_1-03, F16M-023, MEKБ23, F11OB-071]

Image processing in the design and production of IS.pdf (monograph)