Orchard-to-Export: VGG-16 Transfer Learning for Date Fruit Inspection and Quality Grading
Download| Volume 6 Issue 2, 2025 | |
|---|---|
| Author(s): |
Nizamuddin Maitlo Institute of Computer Science, Shah Abdul Latif University, Khairpur, nizamuddin.cs@gmail.com Hidayatullah Shaikh Institute of Computer Science, Shah Abdul Latif University, Khairpur, hidayat.shaikh@salu.edu.pk Atique Ur Rehman Institute of Computer Science, Shah Abdul Latif University, Khairpur, attiqueurrehman.cs@gmail.com Basit Raza Institute of Computer Science, Shah Abdul Latif University, Khairpur, basit.dharejo@salu.edu.pk Fayaz Ahmed Institute of Computer Science, Shah Abdul Latif University, Khairpur, liaquatfayaz0@gmail.com |
| Abstract | Reliable grading of date fruits is essential for ex- port pricing and compliance but still relies largely on human graders, yielding variable quality, limited throughput, and weak traceability. We propose a compact end-to-end computer-vision pipeline for variety identification and quality grading using VGG-16 transfer learning. Experiments use the public Date Fruit Dataset for Inspection and Grading (v3) with four varieties (Aseel, Fasli Toto, Gajar, Kupro) organized by size and grade, captured under controlled illumination. Our training recipe applies light augmentation, ImageNet normalization, optional class-balanced sampling, and partial unfreezing of the last VGG block; optimization uses Adam (10−4), batch size 32, early stopping, and cosine-annealing restarts. On a 70/15/15 stratified split, the held-out test set (256 images) yields 98% accuracy with strong per-class performance (F1 ≥ 0.97 for Aseel and Gajar), with minor confusions between Fasli Toto and Gajar. Learning curves stabilize by epoch 5 without overfitting, and qualitative grids show consistent predictions across sizes and grades. We also outline deployment guidance (illumination control, periodic color calibration, batched real-time inference) and human-in- the-loop verification to support traceability and active learning. Our contributions are an orchard-to-export pipeline, a simple reproducible training recipe for modest datasets, and confusion- aware analyses that surface operational failure modes. |
| Keywords | Date fruit, post-harvest inspection, grading, transfer learning, VGG-16, industrial vision, quality control. |
| Year | 2025 |
| Volume | 6 |
| Issue | 2 |
| Type | Research paper, manuscript, article |
| Recognized by | Higher Education Commission of Pakistan, HEC | Category | Journal Name | ILMA Journal of Technology & Software Management | Publisher Name | ILMA University | Jel Classification | -- | DOI | - | ISSN no (E, Electronic) | 2790-590X | ISSN no (P, Print) | 2709-2240 | Country | Pakistan | City | Karachi | Institution Type | University | Journal Type | Open Access | Manuscript Processing | Blind Peer Reviewed | Format | Paper Link | https://ijtsm.ilmauniversity.edu.pk/arc/Vol6/i2/pdf4.pdf | Page | 64-71 |