Steel-plate surface fault detection

Steel-plate surface fault detection

SAURABH GHATNEKAR

SAURABH GHATNEKAR

Pune, Maharashtra

Aim of this project was to detect and classify surface faults of steel-plates using RPi. Domain: Image Processing,Machine Learning, IoT.

Internet of Things

  • 0 Collaborators

  • 0 Followers

    Follow

Description

Complete project was implemented using Python 3.4 in conda virtual environment. Libraries used: OpenCV , Scikit Image, Scikit Learn

Project was divided into three parts: 1. Image Processing and feature extraction 2. Machine Learning(ML) 3. Raspberry Pi / IoT

1A. Image Processing: Image was captured using USB webcam. Pre-processing included median and Gaussian filtering for noise removal. Then the color image was then converted to grayscale image for feature extraction. this process used OpenCV library

1B. Feature Extraction: From the extensive literature survey it was concluded that Gray level co-occurrence matrix (GLCM) features should be used for ML. The Scikit Image library contains functions for extracting features of a gray image. 5 GLCM properties were extracted : 'contrast','dissimilarity','homogeneity','energy','ASM'. and stored in an excel sheet. Along with this another excel sheet containing labels of image was prepared.

  1. Machine Learning : There are different machine learning algorithms which are used to classify images into different classes. The famous machine learning algorithms are SVM, KNN and Adaboost etc. this system used SVM for roubustness. The Scikit Learn library contains all ML algorithms. Feature Reduction: The ML model was over trained when all five features were for training resulting in false classification. To improve the accuracy only 'contrast','dissimilarity','homogeneity' were used in final training. The accuracy was over 90%

  2. Raspberry Pi. IoT : A green LED indicated Non defective plate and a red LED indicated a defective plate. The results were sent via way2sms API to a stored contact.

Gallery

Links

Partial implementation data

Medium 20621246 1596653140379913 3028956562876983183 n

SAURABH G. created project Steel-plate surface fault detection

Medium 591b1e62 9ded 4ad8 a6f7 9af0b54950ba

Steel-plate surface fault detection

Complete project was implemented using Python 3.4 in conda virtual environment. Libraries used: OpenCV , Scikit Image, Scikit Learn

Project was divided into three parts: 1. Image Processing and feature extraction 2. Machine Learning(ML) 3. Raspberry Pi / IoT

1A. Image Processing: Image was captured using USB webcam. Pre-processing included median and Gaussian filtering for noise removal. Then the color image was then converted to grayscale image for feature extraction. this process used OpenCV library

1B. Feature Extraction: From the extensive literature survey it was concluded that Gray level co-occurrence matrix (GLCM) features should be used for ML. The Scikit Image library contains functions for extracting features of a gray image. 5 GLCM properties were extracted : 'contrast','dissimilarity','homogeneity','energy','ASM'. and stored in an excel sheet. Along with this another excel sheet containing labels of image was prepared.

  1. Machine Learning : There are different machine learning algorithms which are used to classify images into different classes. The famous machine learning algorithms are SVM, KNN and Adaboost etc. this system used SVM for roubustness. The Scikit Learn library contains all ML algorithms. Feature Reduction: The ML model was over trained when all five features were for training resulting in false classification. To improve the accuracy only 'contrast','dissimilarity','homogeneity' were used in final training. The accuracy was over 90%

  2. Raspberry Pi. IoT : A green LED indicated Non defective plate and a red LED indicated a defective plate. The results were sent via way2sms API to a stored contact.

No users to show at the moment.

No users to show at the moment.