Did you know not all pineapples are the same? Just like mango, banana and durians have different varieties.
🍍 Pineapples have distinct cultivars (like Moris, MD2, Josapine and more) each with unique traits and market values. Developing NanasVision to identify them automatically using computer vision has been the highlight of my degree journey!
🔁The Pivot & Field Data Collection:
I initially set out to work on banana disease detection, but after valuable pomology insights from Dr. Aiman Takrim (UPM) on the challenges of single-sensor disease diagnosis, I pivoted to Pineapple Cultivar Classification.
Because no public labeled dataset existed, I had to venture on-site. Datin Suriliza and the Lembaga Perindustrian Nanas Malaysia (LPNM) team connected me with farmers across Melaka and Selangor, allowing me to visit 4 farmlands to collect field data for 4 main cultivars: Moris, MD2, Josapine, and Yankee.
🧠The AI Pipeline & Benchmarking:
Cascaded Architecture: Implemented a two-stage pipeline using YOLOv8n to first detect pineapples, then feeding the image into EfficientNetB0 for cultivar classification.
Model Benchmarking: Evaluated 3 distinct CNN architectures: ResNet50, EfficientNetB0, and MobileNetV3 across augmented and non-augmented data.
30-Test-Case Evaluation: Stress-tested against 30 positive and negative controls. Interestingly, the YOLO detector produced false positives on durians due to high visual similarity in spiky rind textures! I also found that EfficientNet without augmentation suited best for my dataset.
Presentation & FYP 2 Roadmap:
The presentation went smoothly, and my evaluator provided great insights to enhance the system for FYP 2:
1️⃣Upgrading YOLO from binary classifier to multi-class classifier with common fruits to eliminate edge-case false positives (like durians).
2️⃣Expanding the dataset to include harvested pineapples to accommodate real-world consumer and marketplace use cases.
3️⃣Improve UI/UX and add more functionality to the app.
Heartfelt Appreciation:
A massive thank you to my supervisor, Ts. Dr. Norfadzlia, for guiding my AI architecture design, Dr. Md Aiman Takrim Zakaria (UPM) for the domain expertise, Datin Hajah Suriliza, LPNM, & local farmers Pn. Suri, En. Najmi, and En. Nor for the incredible support and field access.
#FinalYearProject #DeepLearning #ComputerVision #YOLOv8 #EfficientNet #MobileNet #AgriTech #LPNM #UTeM #AIArchitecture
17 Agosto 2026 à 02h23
Ver en LinkedIn →
