Session 4: Count Plants using a Classification Model and Geotag them

In Session 4, we’ll teach our AI to spot mango plants in photos using a MobileNet-powered classification model - then take it a step further by tagging each plant’s exact location on the map. By the end, you’ll see an accurate plant count come to life, right where they grow!"

We’ll implement a transfer learning workflow, starting with a MobileNet model pre-trained on ImageNet and fine-tuning it to perform binary classification — mango plant vs. non-mango plant. You’ll understand how feature extraction works in convolutional neural networks (CNNs), how we freeze and unfreeze layers, and why this drastically cuts down training time while boosting accuracy on small datasets.

Once our classification pipeline is running, we’ll integrate it with metadata from geotagged imagery. This means pairing each prediction with GPS coordinates, enabling geo-referenced plant mapping. We’ll explore how to store these coordinates, visualize them on an interactive map, and create a dataset that can support spatial analysis for farm management.

By the end of the session, you’ll have built a full-stack AI + geospatial solution: image classification to detect plants, and geotagging to map them in the real world. This approach is not only useful for agriculture, but also applicable to domains like environmental monitoring, forestry, and asset tracking.