Loading...
Airventra:
Alawlaqi, Noor Almashharawi, Maha Alsalamah, Mashael
Alawlaqi, Noor
Almashharawi, Maha
Alsalamah, Mashael
Citations
Altmetric:
Type
Supervisor
Subject
Date
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Modern warehouses face growing pressure to operate e"ciently while maintaining
inventory accuracy and minimizing human error. This project presents AirVentra, a
smart warehouse management system designed to optimize batch storage location rec ommendation, automate inventory scanning, and streamline task delegation. The system
integrates multiple AI-driven modules including computer vision, topic modeling, and
real-time decision support. Barcode detection is achieved using a YOLOv11n-based
model, which demonstrated superior accuracy (precision: 0.966, recall: 0.967, mAP50:
0.980) and faster inference time (0.9 ms) compared to earlier alternatives. For intelligent
storage location recommendation, the system employs a BERTopic classification model
that semantically categorizes batches based on their textual descriptions. BERTopic out performed LDA and NMF with an overall classification accuracy of 97% and an F1-score
of 0.97 across categories. A dual-phase inventory scanning process enables employees
to first verify batch placement and then assess shelf-level conditions such as overfill, un derfill, or expired items. Detected issues automatically trigger task assignments, with the
system selecting the least-loaded employee to handle the resolution. To support future
automation, we conducted experiments on drone path planning using several algorithms.
The Genetic Algorithm produced the most e"cient route (584.77 meters), outperform ing Christofides, Nearest Neighbor, and Spanning Tree approaches in traversal time. A
web-based system was developed using Laravel and Flask to manage scanning pipelines,
user interfaces, and real-time alerts. Results demonstrate that combining machine learn ing, computer vision, and automation within a unified system can significantly improve
warehouse reliability and reduce manual workload. Future work will expand toward full
autonomous UAV-based scanning and adaptive retraining of models based on evolving
batch metadata and operational patterns. This project is built upon the foundational work
presented in our accepted paper A.1 at the International Symposium on Data Intelligence
and Applications (ISDIA-2025). The paper, which outlines our initial methodology and
experimental setup, is scheduled to appear in Springer’s Lecture Notes in Networks and
Systems (LNNS) series by November 2025