Vladimir Derkachev

RESEARCH, TEACHING, or OTHER INTERESTS

Electrical and Electronic Engineering, Control and Systems Engineering

Scopus Publications

Connected Component Labeling algorithm in streaming image processing with FPGAs
Alexey Bakumenko, Valentin Bakhchevnikov, Vladimir Derkachev, Andrey Kovalev, Vladimir Lobach, and Michael Potipak
SPIE
The task of image fragments extraction with objects of interest is not new. The choice of appropriate segmentation algorithm and its implementation in real time mode become a problem in systems operating with a high-speed video data stream. Connected components labeling is an important step in training samples images preparation for usage in neural networks. To ensure high speed and feasibility of image labeling algorithm for the FPGA, it is necessary to justify the choice of segmentation algorithms, considering the hardware capabilities of the platform. In this paper, we propose a modified one-pass image labeling algorithm for FPGAs, as well as its implementation using the Xilinx System Generator for DSP and the Matlab/Simulink package. As a hardware platform the FMC-200-A mezzanine is used to provide high-speed video data stream from the line camera TELEDYNE DALSA LA-CC-04K05B00-R to the ZYNQ Ultrascale + MPSoC ZCU106 evaluation kit board. The procedure of using hardware implemented SPI interface on the MPSoC ZCU106 board, which is to configure and control the FMC200-A module is described. Implementation of SPI interface is made by using Vivado and Vitis IDE. The labeling results of proposed algorithm on test images, as well as images obtained experimentally are presented.


Remote sensing of agricultural crops seeds for size determination within radar technology
Alexey Bakumenko, Valentin Bakhchevnikov, Vladimir Derkachev, Andrey Kovalev, Vladimir Lobach, and Michael Potipak
SPIE
The high quality of seed cleaning is one of the factors in increasing the yield of agricultural crops. To improve its quality, it is important to estimate average size of seeds for adjusting their feeding method into analysis zone of the optical seeds sorter (the color sorter). Optical methods for average seed size measuring have a significant drawback, which consists in usage of video camera, which lens must be periodically cleaned from settling dust. An alternative method for estimating size of crop seeds is the use of millimeter-wave radar. To assess the geometric characteristics of seeds, it is proposed to use adapted algorithms for sea surface remote sensing. The possibility of applying remote sensing algorithms to the seed layer is justified due to the similarity of geometric structure of the seed layer with a quasi-periodic surface, which is the sea surface. This article discusses an approach for seeds layer spatial model obtaining. The description of backscattered radar signal model is given. We carried out the modeling data analysis and a comparison of estimated average seed size with the geometric characteristics of the layer for different crops. Obtained model results allow us to conclude about the applicability of remote sensing algorithms for geometric characteristics of seeds estimation. Usage of radar data acquisition methods allows to use the available in market mmWave solutions to reduce the cost of measuring equipment for geometric characteristics of seeds for various crops.


Crop seed classification based on a real-time convolutional neural network
Alexey Bakumenko, Valentin Bakhchevnikov, Vladimir Derkachev, Andrey Kovalev, Vladimir Lobach, and Michael Potipak
SPIE
The yield of agricultural crops is directly related to the quality of the seed. One of the ways to improve the quality is to sort seeds by physical and optical properties (color, shade of color, shape, size and structure) using grain cleaning units (photoseparators or color sorters). Currently, two main approaches are used to create sorting algorithms: generalized statistical and neural networks. The benefits of a neural-based sorting approach include the absence of any restrictions on training series; there is no need to pre-examine the nature of the data and adjust sorting algorithm. Nowadays, neural networks are better than statistical methods, especially in the case of deep neural networks. The disadvantages of neural networks are in the high computational complexity of obtaining results and huge size of the training series, sufficient to achieve high accuracy in making reliable decisions. This work discusses the creation of a convolutional neural network architecture that allows dividing the input flow of wheat seeds into two classes: "good" and "bad" (with flaws in shape and color). The implementation of convolutional neural network on FPGA using the Xilinx System Generator for DSP and Matlab / Simulink package is also considered. The results of a convolutional neural network training, assessment of the accuracy of seed correct classification, and verification of the convolutional neural network hardware FPGAs implementation are presented.