|
Machine vision systems used in modern industrial complexes, based on the analysis of multi and hyperspectral imaging. The transition to implementing the "Industry 4.0" program is not possible when using one type of data. The first control system used only the visible range image. They made it possible to analyze the trajectories of movement of objects, control product quality, carry out security functions (control of perimeter crossing), etc. The development of new industrial robotic cells and processing complexes using cognitive functions implying the receipt, analysis, and processing of heterogeneous data. The construction of a unified information field, which allows performing multidimensional operations with data, allows increasing the speed of decision-making and the implementation of automated robot-human systems at the level of an assistant working in a unified workspace. The use of machine vision systems analyzing information received in: visible (shape, the trajectory of movement, position of objects, etc.); near-infrared range (data is similar to visible, allows operation in dusty, foggy, low light conditions); far-infrared range - thermal (plotting temperature gradients, identifying areas of overheating); ultraviolet range (analysis of ionization sources, corona discharges, static charges, tags); X-ray and microwave ranges (analysis of the surface and internal structure of objects, allow the identification of defects); range and 3D sensors (construction of volumetric figures, analysis of the relative position of objects and their interaction), etc. Data analysis is often performed not by a single camera but by a group of sensors located not in a single housing. Primary data integration reduces the number of information channels while maintaining the functionality and accuracy of the analysis. The article discusses creating fusion images obtained by industrial sensors into a combined image containing joint data. Combining multi and hyperspectral imaging makes it possible to increase existing systems' efficiency and implement automated decision-making through their small reconfiguration. The article deals with searching for transformation matrices to create single combined images. A method for forming areas of significance obtained based on the analysis of various channels is proposed As methods of primary data processing, a multicriteria filtering method with an automated selection of processing parameters was used, based on the simultaneous minimization of the L2 norm and the first-order finite differences between the input implementation and the obtained values. The proposed method allows preserving the boundaries of objects and minimizing the noise component both on smooth local sections and near transitions. The transformations of color ranges are carried out using a modified multirange alfa-routing algorithm. The paper proposes an algorithm for fusion images with different coefficients and a criterion for their change in given local areas. On a set of test data obtained by cameras in visible (1024x1024 pixels, 8 bits), nearinfrared (800x600 pixels, 8 bits), thermal imaging (320x240, 8 bits), and depth maps (1024x1024, 8 bits in grayscale), presented examples of the formation of object masks and creating combined images.
|
