Wyniki wyszukiwania - BazTech

1

Defect detection in battery electrode production using supervised and unsupervised learning with laser speckle photometry data

Klarák Jaromír, Chen Lili, Cikalova Ulana, Malík Peter, Andok Robert, Bendjus Beatrice

Advances in Science and Technology. Research Journal

|

2025

|

Vol. 19, no. 4

258--269

EN

The direction for zero-emission transport is largely driven by the adoption of battery-powered vehicles. A critical aspect of a successful battery storage system is the production of high-quality electrodes, which necessitates rigorous inspection processes and defect detection systems. In this paper, we present data obtained using Laser Speckle Photometry (LSP) technology and perform defect detection using two approaches: the YOLOv4 model and the newly developed U2S-CNNv2 model. The U2S-CNNv2 model combines unsupervised and supervised learning to identify defects beyond the training dataset. Our goal is to develop an efficient detection of defects for battery electrode production to meet stringent quality control standards. Our findings show that YOLOv4 is highly effective for deployment in inspection processes, capturing very small defects and operating at 50 frames per second (fps). YOLOv4 achieved an impressive 93.82% accuracy in correctly detecting and 91.10% in correctly labeling defects. Conversely, the U2S-CNNv2 model excels in precisely localizing defect areas and identifying unknown defects or patterns not included in the training dataset. However, it operates at a slower pace of around 3 fps and has a detection accuracy of 83.83% and correct labeling rate of 54.84%.

2

StarCraft Strategy Classification of a Large Human versus Human Game Replay Dataset

Krištofík Štefan, Malík Peter, Kasáš Matúš

Annals of Computer Science and Information Systems

|

2021

|

Vol. 25

137--140

EN

Real-time strategy games are popular in AI research and education. Among them, Starcraft: Brood War (SCBW) is particularly well known. Recently, the largest known SCBW game replay dataset STARDATA was published. We classify player strategies used in the dataset for all 3 playable races and all 6 match-ups. We focus on early to mid-game strategies in matches less than 15 minutes long. By mapping the classified strategies to replay files, we label the files of the dataset and make the labeled dataset available.

3

StarCraft agent strategic training on a large human versus human game replay dataset

Krištofík Štefan, Malík Peter, Kasáš Matúš, Neupauer Štefan

Annals of Computer Science and Information Systems

|

2020

|

Vol. 21

391--399

EN

Real-time strategy games are currently very popular as a testbed for AI research and education. StarCraft: Brood War (SC:BW) is one of such games. Recently, a new large, unlabeled human versus human SC:BW game replay dataset called STARDATA was published. This paper aims to prove that the player strategy diversity requirement of the dataset is met, i.e., that the diversity of player strategies in STARDATA replays is of sufficient quality. To this end, we built a competitive SC:BW agent from scratch and trained its strategic decision making process on STARDATA. The results show that in the current state of the competitive environment the agent is capable of keeping a stable rating and a decent win rate over a longer period of time. It also performs better than our other, simple rule-based agent. Therefore, we conclude that the strategy diversity requirement of STARDATA is met.

4

Instance Segmentation Model Created from Three Semantic Segmentations of Mask, Boundary and Centroid Pixels Verified on GlaS Dataset

Malík Peter, Krištofík Štefan, Knapová Kristína

Annals of Computer Science and Information Systems

|

2020

|

Vol. 21

569--576

EN

Segmentation is the key computer vision task in modern medicine applications. Instance segmentation became the prevalent way to improve segmentation performance in recent years. This work proposes a novel way to design an instance segmentation model that combines 3 semantic segmentation models dedicated for foreground, boundary and centroid predictions. It contains no detector so it is orthogonal to a standard instance segmentation design and can be used to improve the performance of a standard design. The presented custom designed model is verified on the Gland Segmentation in Colon Histology Images dataset.