Mastura Omar

Mastura Omar

Auto-generated profile, based on publicly available information

Mastura Omar

Department of Electrical and Electronics Engineering, Faculty of Engineering, Universiti Putra Malaysia

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

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

Representative Publications

View All Publications
9 September 2026 in Journal of Intelligent Media Computing
Verified
Journal Article
Closed access
MOMastura Omar
Mastura Omar
NKNoor Ain Kamsani
Noor Ain Kamsani
RSRoslina Mohd Sidek
Roslina Mohd Sidek
FRFakhrul Zaman Rokhani
Fakhrul Zaman Rokhani
Object detection models trained on static images may perform poorly when deployed on continuous video, creating challenges for automated industrial inspection. This study investigates whether supplementing static training data with consecutive video frames can improve real-time halal compliance detection in poultry processing. Two YOLOv11l models were trained for 100 epochs under identical hyperparameters. A total of 1,835 video frames were extracted and annotated, with 1,479 allocated to training, 182 to validation and 174 to testing. The baseline model used 1,400 static images, while the proposed model used 2,868 images comprising 1,389 static images and 1,479 video frames. Both models were evaluated on an independent video containing 1,820 frames and 2,439 annotated instances. Although validation performance was similar, video evaluation showed a clear difference. The proposed model increased recall from 49.77% to 60.07% and AP@0.5 from 45.74% to 58.37%, while precision remained comparable at 83.14% and 83.21%, respectively. The proposed model recovered 251 additional structures at the cost of 52 additional false positives and achieved a mean inference time of 9.2 ms per frame. Error analysis showed that most remaining missed detections occurred under severe motion blur, blood coverage, partial visibility or unfavourable orientation. These findings indicate that video-supplemented training improves deployment performance without architectural modification, while human verification remains necessary for final certification.