Food Classification and Quality Inspection

Food Classification and Quality Inspection  System using Deep Convolutional Neural Network

The proposed system performs computer vision task called Food Classification and Quality Inspection in different public/private Food Seller Purchaser Organizations. Food Classification and Quality Inspection system is web application using pre-trained VGG-16 Deep learning classification model. This system is a cross platform which is built on different tools and technologies.

This system is divided into two main modules:-

Food Classifier

Food Classification plays a vital role for the Food Seller and Buyers Organization. In our Ensemble of Deep networks for the classification of Food from Fruit images using the FID30 DataSet. These images are taken at different times of the day and different times of the year. Additionally, the images are taken from a different angle, scale and resolution.

All individual networks takes preprocessed images of 224 x 224 x 3 and batch normalization of 25 images in training and 10 images in testing phase. During training, Networks = {A, B, C, D, E, F, G, H, I, J, K, L, M, N, O, P, Q, R, S, T, U, V, W, X, Y, Z, A1, B1, C1, D1} trained independently. By doing so each network learns its own representation of the image. During testing, the image is passed through the individual networks, and the weighted average of the softmax output gives us the final predictions.

Prediction

Weighted Cross-Entropy loss function

In FCAQI (Food Classification and Quality Inspection) our main motive is to automatic classifying fruits and checking fruit Quality. In FCAQI Food Classification and Quality Inspection Classifier and Inspector do not have an equal amount of training data because all fruits not in single geographical location. In order to handle this unbalanced nature of training data, we use a weighted cross-entropy loss function. It is desirable to give more weight to classes that convergence. Rectified linear units (ReLU) are used as non-linarites.

Summary of Individual Networks

Dataset

We performed our experiments on the FID30 dataset where the images are obtained from real world. The total numbers of training images in the dataset for the classification are classis wise described in given below bar chart:

                                                     Dataset Statistics

Dataset Pre processing

All the classes in our dataset have unbalanced set of images which were resized to maintain standard ratio such that the shorter side has length of 112 and shorter width of 131 is scaled to manage standard required input of our VGG-16 model. The image preprocessing is done using PIL (Python Imaging Library), where each image is read and resized 224 x 224 x 3 because of RGB images.

Image Pre-Processing

Responsive Web Application

For easy to use and fully shared system requirements, we develop a web application rather than develop Desktop, Mobile or Console application. Our web application is developed on Asp.net MVC using different technologies on Server side and Client side.  FCAQI (Food Classification and Quality Inspection) web application is divided into two main modules, which are described below:

Responsive Web Application

Client Side

We use HTML 5, JavaScript and CSS 3 for building responsive web pages. Rather than implement these

Client Side

Technologies directly we use many frameworks such as Bootstrap for responsive ness of web pages and Jquery for making these web pages dynamic and validation of forms. For constructing FCAQI web pages we use Visual Studio as IDE (Integrated Development Environments) which facilitate for RAD (Rapid Application Development).

Server Side

We use Asp.net as a server side framework which receive and process the client requests at server. Requests are managed at server using MVC (Model View Controller) architecture. Controller is coordinator and manager between client Request and Response. We deployed FCAQI (Food Classification and Quality Inspection) on IIS (Internet Information Server). Server side of web application we construct 4 controllers for different purposes, controller names and their purposes are described below:

Server Side

At server side, FCAQI main controller called Model which perform Interprocess communication (IPC) between FCAQI Classifier and Web App for doing Food classification and Quality Inspection. 

Model Controller

Database

          A database is a collection of related data stored in an efficient manner which can be accessed very easily and quickly. In other words database can contain the data about a specific project and metadata. It can be describe the self-describing collection of integrated records.

            The database of proposed system contains the nine table tblSearch, tblVideo, tblWebCam, tblUser, tblContact, tblAboutUs, tblSlider and tblImage.

Class Diagram (CD of FCAQI)

Use Case Diagram

                                        Use Case Diagram FCAQI

Flow Diagram

                                                  Flow Diagram FCAQI

Model Source Code Description                                                           Back To Homepage