Hybrid Deep Learning and Feature Detection for Locating Multiple Circular Areas

This repository explores an approach that combines deep learning with conventional feature detection techniques to locate multiple circular areas in an image.

Problem Formulation

Given a grayscale image containing $N_c$ circular bright spot areas, where:

• $N_c$ is a random number greater than 1.
• The circular areas vary in brightness.
• Their positions are randomly distributed within the image domain.
• The image may contain background noise.

The objective is to develop a model that, when given an input image, accurately returns a list of coordinates corresponding to all detected circular areas.

Model Construction Approach

The process of locating circular areas consists of two key steps:

1. Image Denoising and Brightness Normalization
   A deep convolutional neural network (CNN) is utilized to denoise the image and transform circular regions of varying brightness into multiple circles with nearly uniform brightness.

2. Center Coordinate Extraction
   The Circle Hough Transform is applied to extract the coordinates of the centers of these multi-circles.

Illustration of the model for detecting multiple circular areas

The following presents the training history and evaluation results for the model configured in config.yml:

The Training and Validation Loss over Epochs

Metric	Non-Noised Test Data	Noised Test Data
Total Circles	3,124	3,079
Detected Circles	3,048	2,975
Missed Circles	76	104
False Positives	0	0
Detection Rate (%)	97.57	96.62
Precision	1.00	1.00
Recall	0.98	0.97
F1 Score	0.99	0.98

Circular Area Detection Performance

Getting Started

Requirements

The code has been successfully tested with the following dependencies:

• Python 3.9.21
• TensorFlow 2.18.0
• PyYAML 6.0.2
• OpenCV-Python 4.11.0.86
• NumPy 2.0.2
• Matplotlib 3.9.4

Ensure all required packages are installed before running the code.

Clone the Repository

To begin, clone the repository:

git clone https://github.com/chenbei102/cb_multi_circle_detection.git
cd cb_multi_circle_detection

Usage

Configure Settings (Optional)

The config.yml file allows you to customize various settings. You can modify:

• The datasets used for training, validation, and testing.
• The deep CNN architecture.
• The noise level added to input images.
• Other relevant parameters.

Edit this file as needed before proceeding.

Generate Datasets

Run the following command to generate three datasets for training, validation, and testing:

python gen_dataset.py

Train the Deep CNN Model

python train.py

The trained model and training history will be saved.

Visualize Training History (Optional)

To plot training and validation loss over epochs, run:

python plot_history.py

Evaluate the Model

Assess the model's circular area detection performance with:

python evaluate.py

Evaluation metrics such as detection rate, precision, recall, and F1-score will be calculated.

Visualize Circular Area Detection (Optional)

python plot_predict.py

A sample image will be processed to detect circular regions. The detected circles will be visually overlaid on the original image, which will then be displayed.