gowtham-369/indiancurrencynotesdetection

Purpose and Runtime

This is a Python/Flask web application that detects Indian currency-note denominations in uploaded or camera-supplied images using a custom-trained YOLOv5 model. It returns detection results to a browser UI, which presents the value and produces client-side spoken output for accessibility. Primary runtime entry point: app.py.

HTTP Application Boundary

app.py is the application orchestrator: it serves the HTML frontend and exposes the inference-facing API consumed by browser JavaScript. The server boundary separates browser request handling from Python/PyTorch model execution.

• app.py

• templates/index.html

• static/index.js

Browser UI and Accessibility

The static frontend accepts image input, sends asynchronous requests to the Flask backend, renders returned detection information, and uses browser speech-synthesis APIs for audio output. Its backend URL is explicitly configured in the JavaScript, making deployment host configuration a frontend concern.

• templates/index.html

• static/index.js

• static/index.css

Detection Adapter

yolo_detection.py is the application-specific inference layer between Flask and the vendored YOLOv5 implementation. It is the key boundary for image preprocessing, invoking model inference, and translating detector outputs into currency-recognition results suitable for the API/UI.

• yolo_detection.py

• app.py

• yolov5/detect.py

Language/Result Conversion

language_conversion.py is a distinct application module for converting recognition output into language-oriented presentation data, likely supporting the app’s user-facing and audio-oriented result flow.

• language_conversion.py

• app.py

• static/index.js

ML Model and YOLOv5 Core

The repository vendors YOLOv5 and includes both generic pretrained weights and trained run artifacts. The custom currency model checkpoint is the likely production inference asset; YOLOv5 depends on PyTorch and handles model construction, device execution, and detection post-processing.

• yolov5/runs/train/exp/weights/best.pt

• yolov5/models/yolo.py

• yolov5/utils/general.py

Model Lifecycle and Stored Artifacts

Training, validation, and detection commands remain in-repo, alongside experiment outputs. These are offline/model-development capabilities rather than the web request path. Detection runs persist annotated media and per-image label files; training artifacts retain weights and metrics.

• yolov5/train.py

• yolov5/val.py

• yolov5/runs/detect/

• yolov5/runs/train/exp/

Inputs, Dependencies, and External Services

Sample denomination and mixed-note inputs reside under test_images/. Runtime dependencies are declared at both application and vendored-YOLO levels. The app requires Python 3.7+, Flask, PyTorch/YOLOv5 dependencies, and a browser with speech synthesis; no database or required external inference service is evident. The README identifies PythonAnywhere as a deployed hosting target and Kaggle as the source dataset.

• requirements.txt

• yolov5/requirements.txt

• test_images/