AI Drone Pineapple Counting SaaS
Budget: $200 – $400 USD
I need an end-to-end solution that lets growers upload aerial JPEG, PNG or TIFF shots taken from drones and immediately receive two key pieces of information: the total number of pineapples visible in each image and the average radius of every fruit detected.
The work breaks down into three tightly-linked components:
1. Model creation
• Build and train a computer-vision model (YOLO, Detect-RON, Mask-RCNN or any comparable deep-learning stack you prefer in PyTorch or TensorFlow) that reliably identifies individual pineapples from top-down drone imagery.
• Include a second head or post-processing step that converts pixel dimensions to real-world radius, using the drone’s altitude and sensor data or a calibration grid that I will supply.
• Deliver a reproducible training pipeline plus the final, version-controlled weights.
2. SaaS web app
• Develop a lightweight web interface where users drag-and-drop their aerial photos.
• On upload, trigger the model, return the count, overlay bounding circles, and list radius statistics (min / max / average).
• Provide user authentication, usage logging, and basic billing hooks so I can plug in Stripe later.
• Deploy to a cloud
3. Acceptance criteria
• ≥90 % mAP on a held-out test set I will share.
• End-to-end turnaround under 30 s for a 20 MP image.
• Clean, well-commented code and a README that lets me redeploy from scratch.
• Pineapple count must be 99% accuracy.
This is a greenfield build, so you’re free to propose the exact framework, but I need something production-ready, not just a research demo. If you’ve already tackled orchard or plantation analytics, that experience will be a huge plus.
The work breaks down into three tightly-linked components:
1. Model creation
• Build and train a computer-vision model (YOLO, Detect-RON, Mask-RCNN or any comparable deep-learning stack you prefer in PyTorch or TensorFlow) that reliably identifies individual pineapples from top-down drone imagery.
• Include a second head or post-processing step that converts pixel dimensions to real-world radius, using the drone’s altitude and sensor data or a calibration grid that I will supply.
• Deliver a reproducible training pipeline plus the final, version-controlled weights.
2. SaaS web app
• Develop a lightweight web interface where users drag-and-drop their aerial photos.
• On upload, trigger the model, return the count, overlay bounding circles, and list radius statistics (min / max / average).
• Provide user authentication, usage logging, and basic billing hooks so I can plug in Stripe later.
• Deploy to a cloud
3. Acceptance criteria
• ≥90 % mAP on a held-out test set I will share.
• End-to-end turnaround under 30 s for a 20 MP image.
• Clean, well-commented code and a README that lets me redeploy from scratch.
• Pineapple count must be 99% accuracy.
This is a greenfield build, so you’re free to propose the exact framework, but I need something production-ready, not just a research demo. If you’ve already tackled orchard or plantation analytics, that experience will be a huge plus.