Aircraft Damage Classification & Captioning

Problem

Manual aircraft-damage inspection is slow; the project explores automated classification and natural-language description of visible damage.

Solution

A pretrained VGG16 classifies dent and crack images, while Salesforce BLIP generates captions and summaries for the same visual inputs.

Architecture

Damage image → VGG16 classification branch + BLIP captioning branch → class and natural-language description.

Technical implementation

Uses the CC BY 4.0 Roboflow Aircraft Damage Detection dataset with dent and crack classes, Keras preprocessing, and Hugging Face Transformers.

Key challenges

Coordinate a discriminative image classifier and a generative vision-language model within one notebook workflow.

Lessons learned

Pretrained vision and multimodal models can address complementary inspection tasks without claiming a production safety system.