The repository contains 3 IPYNB files

• Yolov5
• Other_Approaches
• dataset_creater

Rest other are utils and tools which are used along the code

• This file contains code to main approach used for detection
• NOTE It is highly recomended that you upload and execute this file on Google Colab only

• This code file contains 3 other Approaches which were practised along with main code

• Extracting ROIs out of images and creating dataset manualy can be very tedious task
• This code file contains pragram to which make the process very easy
• All the usage instructions are commented inside the file

**All the 3 files are well commented and code instructions and usage of other files present in this repository is directed in them

This contains the defect_box and defect_type csv files

This is uploaded as AlphaQ.pptx

• It is observed that the defect has different texture than the fabric. This is the basis of this approach.
• Combination of Sobelx and Sobely operations are applied and binary thresholding is done. This seperated out the defect from the fabric. The following figure shows the results\

• Borders of image contains noise, thus center portion of the frame is taken as ROI.
• Image is divided into rectangles and pixel intensity of each is noted. Region with highest pixel intensity is considered to the the defect. The following image shows the same.

Following are the results of this method.

• Since all defects share a common texture which is different from the fabric, feature matching can be applied.
• Points in the image with features similar to the defects are returned

• Again exploiting the advantage that defects are textured different from the fabric, we compute lbp
• That pattern is computed as local binary pattern and a histogram of patterns is formed. Then these histograms are fed to SVM Classifier for classification
• Then while testing, the input image is divided into several rectangular sections and each of which is fed into SVM clalssifier.
• The final ROI is taken as the intersection of all ROIs

Following image depects the same

• YOLOv5, an open sourced object detection algorithm by Ultralytics is used
• Efficient and accurate than other object detection algorithms like rCNN, F-rCNN etc