Andrius Kriščiūnas

Dr. Andrius Kriščiūnas

Kaunas University of Technology, Kaunas, Kauno Miestas, Lithuania
orcid

Research Focus & Metrics

Academic Career Length

11Years (2016-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
6 January 2023 in Mathematics
Verified
Journal Article
Open access
PDF file
VGValentas Gruzauskas
Valentas Gruzauskas
Aurelija BurinskienėAurelija Burinskienė
Aurelija Burinskienė
AKAndrius Kriščiūnas
Andrius Kriščiūnas
The growing food demand, the tendency for organic food, and the adaptation of the e-commerce business model require new food supply chain management approaches. On the one hand, 30% of the world’s produced food is wasted, and CO2 emissions are rapidly growing due to transport. On the other hand, the increasingly complex and dynamic environment is decreasing the effectiveness of food supply chains. Because of these trends, sustainability and resilience are becoming more relevant to food supply chains. Therefore, the objective of this paper is to propose a strategy based on information exchange to improve food quality and decrease the level of CO2 emission in last-mile deliveries of food products. To achieve this goal, an agent-based model of last-mile deliveries was developed. The model simulated traffic flow and traffic accidents as disturbances in the system while measuring the level of CO2 emission and food quality of the network. The simulation compares information sharing between all vehicles in the urban area and without information sharing in four scenarios of the food industry. In practice, information sharing is achieved by using connected vehicle technology. The use of information sharing between vehicles in last-mile delivery processes allows the development of a self-organizing system, which would adapt to disturbances and lead to the development of sustainability in the long run.

23 March 2025 in Drones
Verified
Journal Article
Open access
PDF file
JGJonas Gaigalas
Jonas Gaigalas
LPLinas Perkauskas
Linas Perkauskas
HGHenrikas Gricius
Henrikas Gricius
TKTomas Kanapickas
Tomas Kanapickas
AKAndrius Kriščiūnas
Andrius Kriščiūnas
UAVs are vastly used in practical applications such as reconnaissance and search and rescue or other missions which typically require experienced operators. Autonomous drone navigation could aid in situations where the environment is unknown, GPS or radio signals are unavailable, and there are no existing 3D models to preplan a trajectory. Traditional navigation methods employ multiple sensors: LiDAR, sonar, inertial measurement units (IMUs), and cameras. This increases the weight and cost of such drones. This work focuses on autonomous drone navigation from point A to point B using visual information obtained from a monocular camera in a simulator. The solution utilizes a depth image estimation model to create an occupancy grid map of the surrounding area and uses an A* path planning algorithm to find optimal paths to end goals while navigating around the obstacles. The simulation is conducted using AirSim in Unreal Engine. With this work, we propose a framework and scenarios in three different open-source virtual environments, varying in complexity, to test and compare autonomous UAV navigation methods based on vision. In this study, fine-tuned models using synthetic RGB and depth image data were used for each environment, demonstrating a noticeable improvement in depth estimation accuracy, with reductions in Mean Absolute Percentage Error (MAPE) from 120.45% to 33.41% in AirSimNH, from 70.09% to 8.04% in Blocks, and from 121.94% to 32.86% in MSBuild2018. While the proposed UAV autonomous navigation framework utilizing depth images directly from AirSim achieves 38.89%, 87.78%, and 13.33% success rates of reaching goals in AirSimNH, Blocks, and MSBuild2018 environments, respectively, the method with pre-trained depth estimation models fails to reach any end points of the scenarios. The fine-tuned depth estimation models enhance performance, increasing the number of reached goals by 3.33% for AirSimNH and 72.22% for Blocks. These findings highlight the benefits of adapting vision-based models to specific environments, improving UAV autonomy in visually guided navigation tasks.

Verified
Journal Article
Open access
PDF file
TFTautvydas Fyleris
Tautvydas Fyleris
AKAndrius Kriščiūnas
Andrius Kriščiūnas
VGValentas Gruzauskas
Valentas Gruzauskas
Dalia ČalnerytėDalia Čalnerytė
Dalia Čalnerytė
RBRimantas Barauskas
Rimantas Barauskas
Urban change detection is an important part of sustainable urban planning, regional development, and socio-economic analysis, especially in regions with limited access to economic and demographic statistical data. The goal of this research is to create a strategy that enables the extraction of indicators from large-scale orthoimages of different resolution with practically acceptable accuracy after a short training process. Remote sensing data can be used to detect changes in number of buildings, forest areas, and other landscape objects. In this paper, aerial images of a digital raster orthophoto map at scale 1:10,000 of the Republic of Lithuania (ORT10LT) of three periods (2009–2010, 2012–2013, 2015–2017) were analyzed. Because of the developing technologies, the quality of the images differs significantly and should be taken into account while preparing the dataset for training the semantic segmentation model DeepLabv3 with a ResNet50 backbone. In the data preparation step, normalization techniques were used to ensure stability of image quality and contrast. Focal loss for the training metric was selected to deal with the misbalanced dataset. The suggested model training process is based on the transfer learning technique and combines using a model with weights pretrained in ImageNet with learning on coarse and fine-tuning datasets. The coarse dataset consists of images with classes generated automatically from Open Street Map (OSM) data and the fine-tuning dataset was created by manually reviewing the images to ensure that the objects in images match the labels. To highlight the benefits of transfer learning, six different models were trained by combining different steps of the suggested model training process. It is demonstrated that using pretrained weights results in improved performance of the model and the best performance was demonstrated by the model which includes all three steps of the training process (pretrained weights, training on coarse and fine-tuning datasets). Finally, the results obtained with the created machine learning model enable the implementation of different approaches to detect, analyze, and interpret urban changes for policymakers and investors on different levels on a local map, grid, or municipality level.

15 November 2022 in Agriculture
Verified
Journal Article
Open access
PDF file
RBRimantas Barauskas
Rimantas Barauskas
AKAndrius Kriščiūnas
Andrius Kriščiūnas
Dalia ČalnerytėDalia Čalnerytė
Dalia Čalnerytė
PPPaulius Pilipavičius
Paulius Pilipavičius
TFTautvydas Fyleris
Tautvydas Fyleris
RMRobertas Mikalauskis
Robertas Mikalauskis
Automatic climate management enables us to reduce repetitive work and share knowledge of different experts. An artificial intelligence-based layer to manage climate in white button mushroom growing hall was presented in this article. It combines visual data, climate data collected by sensors, and technologists’ actions taken to manage climate in the mushroom growing hall. The layer employs visual data analysis methods (morphological analysis, Fourier analysis, convolutional neural networks) to extract indicators, such as the percentage of mycelium coverage and number of pins of different size per area unit. These indicators are used to generate time series that represent the dynamics of the mushroom growing process. The incorporation of time synchronized indicators obtained from visual data with monitored climate indicators and technologists’ actions allows for the application of a supervised learning decision making model to automatically define necessary climate changes. Whereas managed climate parameters and visual indicators depend on the mushroom production stage, three different models were created to correspond the incubation, shock, and fruiting stage of the mushroom production process (using decision trees, K-nearest neighbors’ method). An analysis of the results showed that trends of the selected visual indicators remain similar during different cultivations. Thus, the created decision-making models allow for the definition of the majority of the cases in which the climate change or transition between the growing stages is needed.

Verified
Journal Article
Open access
PDF file
AŠArminas Šidlauskas
Arminas Šidlauskas
AKAndrius Kriščiūnas
Andrius Kriščiūnas
Dalia ČalnerytėDalia Čalnerytė
Dalia Čalnerytė
Satellite image generation has a wide range of applications. For example, parts of images must be restored in areas obscured by clouds or cloud shadows or areas that must be anonymized. The need to cover a large area with the generated images faces the challenge that separately generated images must maintain the structural and color continuity between the adjacent generated images as well as the actual ones. This study presents a modified architecture of the generative adversarial network (GAN) pix2pix that ensures the integrity of the generated remote sensing images. The pix2pix model comprises a U-Net generator and a PatchGAN discriminator. The generator was modified by expanding the input set with images representing the known parts of ground truth and the respective mask. Data used for the generative model consist of Sentinel-2 (S2) RGB satellite imagery as the target data and OpenStreetMap mapping data as the input. Since forested areas and fields dominate in images, a Kneedle clusterization method was applied to create datasets that better represent the other classes, such as buildings and roads. The original and updated models were trained on different datasets and their results were evaluated using gradient magnitude (GM), Fréchet inception distance (FID), structural similarity index measure (SSIM), and multiscale structural similarity index measure (MS-SSIM) metrics. The models with the updated architecture show improvement in gradient magnitude, SSIM, and MS-SSIM values for all datasets. The average GMs of the junction region and the full image are similar (do not exceed 7%) for the images generated using the modified architecture whereas it is more than 13% higher in the junction area for the images generated using the original architecture. The importance of class balancing is demonstrated by the fact that, for both architectures, models trained on the dataset with a higher ratio of classes representing buildings and roads compared to the models trained on the dataset without clusterization have more than 10% lower FID (162.673 to 190.036 for pix2pix and 173.408 to 195.621 for the modified architecture) and more than 5% higher SSIM (0.3532 to 0.3284 for pix2pix and 0.3575 to 0.3345 for the modified architecture) and MS-SSIM (0.3532 to 0.3284 for pix2pix and 0.3575 to 0.3345 for the modified architecture) values.

24 August 2022 in Elektronika ir Elektrotechnika
Verified
Journal Article
Open access
PDF file
DKDovile Komolovaite
Dovile Komolovaite
AKAndrius Kriščiūnas
Andrius Kriščiūnas
ILIngrida Lagzdinyte-Budnike
Ingrida Lagzdinyte-Budnike
ABAurelijus Budnikas
Aurelijus Budnikas
DRDominykas Rentelis
Dominykas Rentelis
Vehicle detection and classification is an important part of an intelligent transportation surveillance system. Although car detection is a trivial task for deep learning models, studies have shown that when vehicles are visible from different angles, more research is relevant for brand classification. Furthermore, each year, more than 30 new car models are released to the United States market alone, implying that the model needs to be updated with new classes, and the task becomes more complex over time. As a result, a transfer learning approach has been investigated that allows the retraining of a model with a small amount of data. This study proposes an efficient solution to develop an updatable local vehicle brand monitoring system. The proposed framework includes the dataset preparation, object detection, and a view-independent make classification model that has been tested using two efficient deep learning architectures, EfficientNetV2 and MobileNetV2. The model was trained on the dominant car brands in Lithuania and achieved 81.39 % accuracy in classifying 19 classes, using 400 to 500 images per class.