Automatic Damage Detection

Application for Automatic Damage Detection in Asset Images

AI-powered service for detecting and identifying objects in images, trained to automatically locate various types of damage in civil engineering asset photos, such as moisture stains, cracks, or concrete spalling.

A web application providing an intuitive interface for complex AI models, meticulously trained for each damage type, allowing users to leverage advanced technology without specialized knowledge. It significantly speeds up asset inspections and simplifies report generation by overlaying a damage mask on the analyzed image and providing statistical insights.

Main Features

  • Automatic Mask Generation

    • Machine learning models trained to generate damage masks on images.
    • Automatic identification of selected damage types.
    • Supports images of any size, with automatic cropping for model adaptation.

  • Interactive Validation Tool

    • Cropped images are reconstructed after validation.
    • Unified masks provide a final PNG file with pixel-accurate alignment to the input image.
    • Final statistics detail damage proportions and model specifications.

  • Results Generation

    • Tool for validating generated masks.
    • Users can register masks as correct or edit them by drawing directly on the image.

  • Web Interface

    • User-friendly interface for uploading images, processing them, and obtaining results.
    • Continuous processing status tracking for all uploaded images.
    • Automatic saving allows users to pause and resume work at any time.

    • Damage Mask Generation

    The core of this application is AI-generated damage masks. The models perform three tasks:
    • Locate areas containing "damage objects".
    • Identify the type of damage (moisture, cracks, etc.).
    • Draw an exact mask highlighting the pixels that constitute the identified damage.
    Combining all identified damages results, a comprehensive mask is generated that overlays the analyzed image, visually indicating damaged areas.

    Fine-Tuning

    Using fine-tuning on pre-trained generalist models achieves precise results with shorter training times and significantly smaller datasets than training models from scratch.

    Retraining

    Once validated by experts, masks are stored in a database and used for model retraining, continuously improving accuracy as the models are used more frequently.

    Cloud-Based Architecture

    The back-end is fully cloud-based and predominantly serverless, ensuring total availability, the ability to perform computations on specialized hardware, and easy system scalability.

    Request a demo