Computer Sciences and Information Technology
Title : | A Low Cost, Portable and High Quality Device (Prototype) for Food Quality Assessment Based on Microscopic Image and Deep Learning |
Area of research : | Computer Sciences and Information Technology |
Focus area : | Development of microscopic image capturing device |
Principal Investigator : | Dr Shovan Barma, Assistant Professor, Indian Institute of Information Technology (IIT) Guwahati |
Timeline Start Year : | 2018 |
Contact info : | shovan@iiitg.ac.in; shovan.barma@gmail.com ; anirban@ee.iitkgp.ac.in |
Details
Executive Summary : | This project aims to develop a low-cost portable and high-quality device for food quality assessment systems. To achieve the goal microscopic image will be considered which could provide the detailed assessment than normal images of a food sample. For this purpose, a microscopic image capturing system will be developed. A deep learning-based algorithm will be developed on android platform to analyze the captured microscopic images of food samples. Several microscopic images of the food sample (e.g., fresh and rotten) will be taken using the developed device to train the deep learning network and to assess the quality of the food numerous features including color, texture and other relevant features will be taken into account. The main component the microscopic image capturing system will be developed using “Foldscope”. The Foldscope is an optical microscope (paper microscope) that can be assembled in a small space that includes a spherical glass lens, a LED and a diffuser panel, along with a watch battery that powers the LED. After assembly, it becomes the size of a bookmark and weighs of about 8 grams. The lenses can provide magnification from 140X to 2000X and the images can be captured by a simple smartphone. Most importantly, it costs less than one hundred rupees per piece. Such features of the Foldscope will make the overall microscopic image capturing system portable, low cost and easily usable. Another advantage of the Foldscope is that the captured image can be fed to the android system very easily where the deep learning recognition task and food quality analysis will be accompanied. Therefore, the proposed system will work in the following manner: First, a food sample for test will be prepared following the guidelines of the use of Foldscope. Then the microscopic image will be captured by a simple smartphone camera. Further, the captured image will be analyzed based on CNN based for recognition purposes which perform better than the conventional neural network methods. In this regard, an App on android platform will be developed which will also help to visualize the results and documentation. The training data set will be generated certainly. It is evident that a large scale of data set for network training improves the recognition result. But it is not a trivial task to collect a large amount of data during developing the system as in this device (prototype), investigators intend to use only two types of foods including one vegetable (potato) and one fruit (apple) for a test case. Consequently, the same kinds of vegetables or fruits include several varieties. Therefore, to deal with such a situation transfer learning will be adopted which could improve the system performance. |
Co-PI: | Dr Anirban Mukherjee, Associate Professor, Indian Institute of Technology (IIT) Kharagpur |
Total Budget (INR): | 17,00,873 |
Achievements : | 1) In the first work, an algorithm is developed to enhance the Foldscope Images in HSV Space by PSO Optimization Technique. The work has been published in an International Conference IEEE BMEiCON 2019, Ubon Ratchathani, Thailand, Thailand Nov. 19-22, 2019. 2).
2) Verified the usability of foldscope as a food quality assessment device.
3) Estimated the starch content of potato tuber by using a traditional microscope and Foldscope. The results are very consistent. The work has been published in an international conference PReMI 2019, Tezpur, India, Dec 17-20, 2019. |
Publications : | 4 |
Organizations involved