Skip to content

image identification, whether a mutton meatball is infected by fungal and whether it's adulterated with duck, Build with Baidu PaddlePaddle DL Framework

License

Notifications You must be signed in to change notification settings

castorwong/identifying_mutton_fungal-infection_and_whether_adulterated_with_duck

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

identifying mutton fungal-infection and whether its adulterated with duck

Notice: This Project is a subproject of my senior's MSE degree's Paper (palenn (Zhaochangtong) (github.com)). Salute!

My senior's MSE degree paper's content is identifing whether the mutton meatball is fungal-infected and whether the meatball is adulterated with duck by meanings of the fusion of E-nose and NIR(Near Infrared) features.

So I create this repository , try to do the same identification using only the images of those meatballs.

Requirement:

python=3.7 (Anaconda is recommended)

PaddlePadddle 2.4.2(must be 2.4.x ,pd.dataloader()has been refactored by Baidu since 2.5.0)

Install on Windows via PIP-Document-PaddlePaddle Deep Learning Platform

pip install the packages below:

paddlex

PyQt5

scikit-image

jsonpath

filelock

(Just pip install the packages above, nothing else needed)

Introduction

  1. MainWindow MainWindow

  2. MainWindow + MobileNetV3 Result

MainWindow2

  1. Baidu Paddle AI Studio , Online Predict Result UI

notice: the token has expired, so the online prediction doesn't work now, the HTTP Reqest Help Document can be found at: https://cloud.baidu.com/doc/EASYDL/s/Sk38n3baq With Baidu EasyDL , the prediction model is trained online. The test image is uploaded to the cloud and EasyDL will return a json file. Just split the json file to exact what you want

FYI: the CPU solution on the right side (“4.2 CPU方案”): SIFT (feature) + Bow(bag of words) + SVM(supporting vector machine) & HOG(Histogram of Oriented Gradient) + Kernel SVM have not been implement

DataSet:

all the pics are shot by a Canon EOS 600D , auto exposure and auto white balance

DataSet:https://pan.baidu.com/s/1UjLCDVn1Yed_4pqet-d26g?pwd=data

example_pic

the white balance is not accurate by the cam, but we put a white board in the pic as a pure white reference to calculate the white balance afterwards.

the meatball is placed in a round Petri dish, before inputing into the predict algorithm

, I use Hough Circle Transform to detect the round Petri dish and crop it out.

The Filename example : CK-100%-0 (2).JPG or J-45%-2 (5).JPG :

CK / J : CK means no fungal-infection, J means with fungal-infection

0%~100% : means the mutton meatball is 0%~100% adulterated with duck. For instance, 45% means the mutton meatball is 45% adulterated with duck.

​ I cut 0%~45% as "low" and "60%~100%" as "high"

0~5 : not invoved in this subproject, this number means the meatball were placed in the petri dish for 0~5 days before this photo was shot.

About

image identification, whether a mutton meatball is infected by fungal and whether it's adulterated with duck, Build with Baidu PaddlePaddle DL Framework

Resources

License

Stars

Watchers

Forks