Merge pull request #35 from PUTvision/pprai2025

pprai2025 md added

Author: 2acholsk1

_posts/2025/04/2025-04-11-pprai-conference.md

@@ -0,0 +1,37 @@
+---
+layout: single
+title: "PPRAI 2025"
+author: [zacholski-piotr, piechocki-mateusz, kraft-marek]
+modified: 2025-04-11
+tags: [computer vision, deep learning, robotics, artificial inteligence, remote sensing]
+category: [conference]
+teaser: "/assets/images/posts/2025/04/pprai-2025-logo.webp"
+---
+
+<p align="center">
+    <img src="/assets/images/posts/2025/04/pprai_2025_mp2_1.webp" height="300px" />
+</p>
+
+From April 7th to 9th, 2025, our team of three had the pleasure of attending the 6th Polish Conference on Artificial Intelligence (PP-RAI 2025) in Katowice. The event was hosted at the Silesian Museum and organized by the University of Silesia in Katowice.
+
+## Presentations abstracts
+
+**Anomaly Detection in Ground-Based Sky Imagery**
+
+> Ensuring the reliability of neural networks in industrial applications is challenging due to data drift and anomalies, especially in nonstationary environments, such as solar irradiance forecasting based on sky images. This study presents an image encoder embeddings classifier based on the isolation forest algorithm to detect outliers in data streams. By autonomously monitoring data validity at remote locations, the presented method aims to enhance forecast reliability, minimizing disruptions due to unexpected variations. This approach is a step towards the seamless integration of solar irradiance forecasting models into smart grids and energy management systems, contributing to the more reliable and efficient use of renewable energy.
+
+<p align="center">
+    <img src="/assets/images/posts/2025/04/pprai_2025_mp.webp" height="300px" />
+</p>
+
+**A Simple and Efficient Method for GPS-less Drone Navigation Using Visual Cues**
+
+> Accurate localization on the map is a key feature of geolocalisation and navigation of aerial robots. In this preliminary study, we compare two localisation strategies based on Local Binary Patterns (LBP): (1) random initialisation subsequently processed with particle filter and (2) same as (1), but initialised using a place recognition neural network trained with metric learning. A range of peace recognition models based on ResNet were evaluated to select the highest performing model. The results clearly show that the neural network-based initialisation results in improved localization accuracy and faster particle filter convergence as compared to random sampling, with minimal computational overhead.
+
+<p align="center">
+    <img src="/assets/images/posts/2025/04/pprai_2025_pz_mk_1.webp" height="300px" />
+</p>
+
+PP-RAI 2025 served as an excellent platform for exchanging ideas and engaging with fellow researchers and professionals in the field of artificial intelligence.​
+
+We extend our sincere thanks to the organizers for facilitating such a well-structured event. For more information about the conference, please visit the official website: https://pp-rai.pl/.