CanSat Terrain Image Classification
Python terrain-image processing for Project Trailblazer, with image tiling, a random forest baseline and a transferred ResNet18 classifier.
What did the terrain project do?
The Project Trailblazer CanSat project processed images taken by a probe into terrain categories such as forest, grass and sand. The public repository calls it picture segmentation and describes image classification pipelines.
My role
I developed the image-processing models as a member of the Project Trailblazer team. The repository lists me as the author.
Implementation
Data preparation resized large images and divided them into 224 × 224 pixel tiles. One notebook trained a random forest classifier; another adapted a pretrained ResNet18 by replacing its final dense layer. Inference notebooks loaded images and saved weights to produce the output.
Technologies and evidence
Python notebooks, NumPy, PIL and PyTorch. The repository includes model notebooks and example outputs.