Mr. Ebenezer Olaniyi Profile

Ebenezer Olaniyi

SPIE Involvement:

Author

Publications (1)

Proceedings Article | 13 June 2023 Presentation + Paper

Enhanced segmentation of beef longissimus dorsi muscle using structured illumination reflectance imaging with deep learning

Jiaxu Cai, Yuzhen Lu, Ebenezer Olaniyi, Shangshang Wang, Chelsie Dahlgren, Derris Devost-Burnett, Thu Dinh

Proceedings Volume 12545, 1254506 (2023) https://doi.org/10.1117/12.2663575

KEYWORDS: Image segmentation, Light sources and illumination, Muscles, Spatial frequencies, Deep learning, Education and training, Biological samples, Statistical modeling, Image processing, Image classification

Read Abstract +

Marbling is one of the most important determinants for beef quality and assessed in terms of the abundance and spatial distribution of visible fat flecks in the longissimus dorsi (LD) muscle. Visual appraisal by trained professionals is the currently prevailing practice in beef marbling assessment, but it suffers from the drawbacks of human subjective, laborintensive, and time-consuming. Computer vision technology offers a promising alternative for objective and automated beef quality assessment. Recently, imaging technology under structured illumination has emerged for the detection of meat quality characteristics as opposed to existing imaging modalities using uniform illumination. By modulating light at certain spatial frequencies, structured illumination reflectance imaging has the potential to resolve subtle texture features of meat surface/subsurface for enhanced quality assessment. This study represents the proof-of-concept evaluation of the applicability of an inhouse-assembled structured illumination imaging system, combined with deep learning, for beef marbling assessment. Beef samples of varying marbling degrees were imaged under sinusoidal illumination at spatial frequencies of 0.05-0.40 cycles/mm. The acquired images were demodulated into direct component (DC) and amplitude component (AC) images at each spatial frequency. A deep learning segmentation model, SegFormer, was built using DC and AC (0.05-0.40 cycles/mm) images for segmenting the LD muscle. Texture features were extracted by a pretrained ResNext model from the LD muscle segments, and then used for building discriminative models to classify samples of three marbling categories. The DC images yielded the overall classification accuracy of 76.32%, while the AC images resulted in improved accuracies of 76.84%-81.05%, which the best accuracy attained at the spatial frequency of 0.40 cycles/mm. This study shows the effectiveness of imaging under sinusoidal illumination in place of uniform illumination for enhanced assessment of beef marbling.

View contact details

UPDATE YOUR PROFILE

Is this your profile? Update it now.

Sign into your SPIE.org account

Don’t have a profile and want one?

Create an account on SPIE.org

Keywords/Phrases

Search In:

Publication Years