95%
classification accuracy
Computer vision2023Personal project
PlantVision AI
PyTorch plant species classifier trained across a 40,000-image dataset and evaluated across 25 plant species.
25
plant species
40K+
training images
PyTorch
core stack
Overview
A practical computer vision classifier.
PlantVision AI was a personal research project using deep learning to classify plant species from images. The goal was to demonstrate practical neural-network work in a domain connected to plant discovery and identification.
The project involved processing more than 40,000 plant images across 25 species, implementing a training pipeline, and evaluating model performance against a held-out validation set.
Objectives
What the model needed to prove.
Model accuracy
Create a classifier accurate enough to distinguish visually similar plant categories across a controlled species set.
Large-scale processing
Prepare, normalize, augment, and train against a large image dataset without making the pipeline brittle.
Training performance
Use GPU-accelerated training and experimentation loops to improve model performance efficiently.
Reusable architecture
Keep the approach extensible so more species or an app-facing inference layer could be added later.
Technical approach
How the classifier was built.
CNN architecture
Implemented convolutional neural network patterns suited for image classification, including normalization and regularization techniques.
Transfer learning
Used pre-trained model patterns as a foundation, then fine-tuned against plant image data to improve accuracy with less training overhead.
Data preprocessing
Built preprocessing steps for image resizing, augmentation, normalization, and train/validation splitting to improve generalization.
Validation methodology
Tracked accuracy across validation data to avoid relying on training performance alone and to expose weak categories.
Skills
Tools and concepts demonstrated.
PyTorchDeep learningComputer visionPythonNeural networksTransfer learningGPU training