An insightful analysis of CNN-based dietary medicine recognition
In the quest for precise dietary medicine recognition, i.e. seed classification, this paper profoundly investigates some state-of-the-art deep learning models, namely VGG16, MobileNet, InceptionV3, ResNet50, and a hybrid combination thereof. The research utilizes an extensive data set featuring eigh...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-03-01
|
| Series: | Journal of Agriculture and Food Research |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S266615432400601X |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In the quest for precise dietary medicine recognition, i.e. seed classification, this paper profoundly investigates some state-of-the-art deep learning models, namely VGG16, MobileNet, InceptionV3, ResNet50, and a hybrid combination thereof. The research utilizes an extensive data set featuring eight distinct dietary medicines, including chia seeds, flax seeds, garden cress seeds, hyptis suaveolens, plantago ovata, pumpkin seeds, tragacanth gum, and white sesame seeds, for machine-vision-based recognition. Through meticulous experimentation, our deep-learning modeling demonstrates distinguishing results, revealing the models’ capacity to discern subtle distinctions among various dietary medicines. Supporting evidence, including confusion matrices and training histories, corroborates the impartiality of our training procedures. These collective efforts yield highly competitive performance metrics, with accuracy metrics consistently surpassing 96 % across all deep learning models. Notably, the hybrid or custom-built model, comprising four distinct models ResNet50, VGG16, MobileNet, and InceptionV3 attains an exceptional accuracy of approximately 99.54 %. The cost-sensitive Neural Network approach was used during the training of all the models to achieve similar behavior of a balanced dataset. The hybrid model uses an average ensemble approach. Nevertheless, our unwavering commitment to excellence continues to drive us to explore further refinements and optimizations to augment the resilience and precision of our seed classification models. |
|---|---|
| ISSN: | 2666-1543 |