Image Classification for Solar Panel Fault Detection

Overview

This repository contains the implementation and report of an image classification project focused on fault detection in solar panels. The project includes two major aspects: a robust feedforward neural network for character recognition and a dedicated image classification framework for precise solar panel fault detection. The project leverages machine learning and deep learning techniques to address real-world challenges in renewable energy, quality control, and operational automation.

Table of Contents

• Project Highlights
• Project Structure
• Objective and Applications
• Introduction
• Approach
• Results and Discussion
• Future Modifications
• References

Project Highlights

• Developed a feedforward neural network for accurate character recognition achieving over 89% accuracy.
• Created an advanced image classification framework for detecting faults in solar panels.
• Demonstrated applications in solar panel fault detection, automated form processing, and renewable energy quality control.
• Explored optimization techniques like Adam, RMSProp, and Momentum for enhanced model performance.
• Contributed to the sustainability and efficiency of solar energy systems and quality control processes.

Project Structure

• notebooks/: Jupyter notebooks illustrating data preprocessing, model training, and evaluation.
• results/: Contains visualizations, graphs, accuracy plots, and confusion matrices.
• README.md: Detailed project description and overview.

Objective and Applications

The primary objective of this project is to develop an image classification framework for accurate fault detection in solar panels. The project extends to applications including:

1. Solar Panel Fault Detection and Maintenance: Identify and localize defects to optimize panel performance.
2. Automated Form Processing: Automate data extraction from handwritten forms for efficient data entry.
3. Renewable Energy Quality Control: Assess manufacturing integrity and product quality in solar panels.
4. Real-time Monitoring and Analytics: Enable swift response to potential issues for operational efficiency.

Introduction

The project addresses fundamental challenges in pattern recognition and machine learning, including character recognition and solar panel fault detection. The report showcases the development of a feedforward neural network for character recognition and its extension to detect various types of defects in solar panels.

Approach

1. Character Recognition:

   • Trained a feedforward neural network using the EMNIST dataset for accurate digit and character recognition.
   • Explored data preprocessing, including reshaping, normalization, and one-hot encoding.
   • Designed a neural network architecture with activation functions and optimized hyperparameters.

2. Solar Panel Fault Detection:

   • Created an image classification framework for identifying defects in solar panels.
   • Utilized optimization techniques like Adam, RMSProp, and Momentum for model enhancement.
   • Explored accuracy, confusion matrices, and result visualizations.

Results and Discussion

The character recognition neural network achieved over 90% accuracy on the EMNIST dataset, demonstrating its efficacy in accurate recognition. For solar panel fault detection, the framework showcased promising results with optimized optimization techniques. Result visualizations include accuracy graphs and confusion matrices.

Future Modifications

Future improvements for the project include:

• Experimenting with advanced neural network architectures like CNNs.
• Applying data augmentation techniques for better model generalization.
• Fine-tuning hyperparameters and exploring ensemble methods.
• Enhancing the data preprocessing pipeline for solar panel images.

References

1. Deep Learning by Prof. Mitesh Khapra[http://www.cse.iitm.ac.in/~miteshk/CS6910.html]
2. Linear Algebra by Prof. Gilbert Strang[https://ocw.mit.edu/courses/18-06-linear-algebra-spring-2010/]

---

This project was completed as part of an internship at CSIR-Central Scientific Instruments Organisation (CSIO) during [July - AUgust 2023]. For inquiries, contact [adhithyarg26@gmail.com].