Deep Learning-Based Meat Freshness Detection with Segmentation and OOD-Aware Classification
arXiv:2603.00368v1 Announce Type: new Abstract: In this study, we present a meat freshness classification framework from Red-Green-Blue (RGB) images that supports both packaged and unpackaged meat datasets. The system classifies four in-distribution (ID) meat classes and uses an out-of-distribution (OOD)-aware abstention mechanism that flags low-confidence samples as No Result. The pipeline combines U-Net-based segmentation with deep feature classifiers. Segmentation is used as a preprocessing step to isolate the meat region and reduce background, producing more consistent inputs for classification. The segmentation module achieved an Intersection over Union (IoU) of 75% and a Dice coefficient of 82%, producing standardized inputs for the classification stage. For classification, we benchmark five backbones: Residual Network-50 (ResNet-50), Vision Transformer-Base/16 (ViT-B/16), Swin Transformer-Tiny (Swin-T), EfficientNet-B0, and MobileNetV3-Small. We use nested 5x3 cross-validation
arXiv:2603.00368v1 Announce Type: new Abstract: In this study, we present a meat freshness classification framework from Red-Green-Blue (RGB) images that supports both packaged and unpackaged meat datasets. The system classifies four in-distribution (ID) meat classes and uses an out-of-distribution (OOD)-aware abstention mechanism that flags low-confidence samples as No Result. The pipeline combines U-Net-based segmentation with deep feature classifiers. Segmentation is used as a preprocessing step to isolate the meat region and reduce background, producing more consistent inputs for classification. The segmentation module achieved an Intersection over Union (IoU) of 75% and a Dice coefficient of 82%, producing standardized inputs for the classification stage. For classification, we benchmark five backbones: Residual Network-50 (ResNet-50), Vision Transformer-Base/16 (ViT-B/16), Swin Transformer-Tiny (Swin-T), EfficientNet-B0, and MobileNetV3-Small. We use nested 5x3 cross-validation (CV) for model selection and hyperparameter tuning. On the held-out ID test set, EfficientNet-B0 achieves the highest accuracy (98.10%), followed by ResNet-50 and MobileNetV3-Small (both 97.63%) and Swin-T (97.51%), while ViT-B/16 is lower (94.42%). We additionally evaluate OOD scoring and thresholding using standard OOD metrics and sensitivity analysis over the abstention threshold. Finally, we report on-device latency using TensorFlow Lite (TFLite) on a smartphone, highlighting practical accuracy-latency trade-offs for future deployment.
Executive Summary
This study presents a deep learning-based framework for meat freshness classification from RGB images, combining U-Net-based segmentation and deep feature classifiers. The framework effectively handles both packaged and unpackaged meat datasets, achieving state-of-the-art results on four in-distribution meat classes. Notably, the study incorporates an out-of-distribution (OOD)-aware abstention mechanism, which flags low-confidence samples as No Result. The results demonstrate the framework's practical accuracy-latency trade-offs, with EfficientNet-B0 achieving the highest accuracy (98.10%) on the held-out ID test set. The study's findings have significant implications for the food industry, particularly in ensuring meat quality and safety.
Key Points
- ▸ The framework combines U-Net-based segmentation and deep feature classifiers for meat freshness classification.
- ▸ It effectively handles both packaged and unpackaged meat datasets.
- ▸ The OOD-aware abstention mechanism flags low-confidence samples as No Result.
Merits
Strength in segmentation module
The U-Net-based segmentation module achieved an Intersection over Union (IoU) of 75% and a Dice coefficient of 82%, producing standardized inputs for the classification stage.
State-of-the-art results on in-distribution meat classes
EfficientNet-B0 achieved the highest accuracy (98.10%) on the held-out ID test set, followed by ResNet-50 and MobileNetV3-Small (both 97.63%).
Practical accuracy-latency trade-offs
The study reports on-device latency using TensorFlow Lite (TFLite) on a smartphone, highlighting the framework's practical accuracy-latency trade-offs for future deployment.
Demerits
Limitation in OOD scoring and thresholding
The study evaluates OOD scoring and thresholding using standard OOD metrics, but the sensitivity analysis over the abstention threshold is limited.
Potential bias in dataset
The study may be biased towards the specific dataset used, which may not be representative of all meat freshness classification scenarios.
Expert Commentary
The study presents a robust and effective framework for meat freshness classification from RGB images. The combination of U-Net-based segmentation and deep feature classifiers is a key strength of the framework. However, the study's limitation in OOD scoring and thresholding is a concern, as it may impact the framework's performance in real-world scenarios. Additionally, the potential bias in the dataset used may limit the study's generalizability. Nevertheless, the study's findings have significant implications for the food industry, and the framework's practical accuracy-latency trade-offs make it an attractive solution for future deployment.
Recommendations
- ✓ Future studies should focus on addressing the limitation in OOD scoring and thresholding, as well as exploring the framework's generalizability to different meat classes and datasets.
- ✓ The food industry should consider adopting the framework for on-the-go meat freshness classification, particularly in high-risk scenarios such as meat processing and handling.
