Source: ALABAMA A&M UNIVERSITY submitted to
A PORTABLE WEIGHT APPROXIMATION SYSTEM FOR SWINE PRODUCTION USING MACHINE VISION AND NEURAL NETWORKS
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
TERMINATED
Funding Source
NRI COMPETITIVE GRANT
Reporting Frequency
Annual
Accession No.
0197354
Grant No.
2003-35503-13990
Project No.
ALAX-012-203
Proposal No.
2003-01517
Multistate No.
(N/A)
Program Code
71.1
Project Start Date
Sep 15, 2003
Project End Date
Sep 14, 2005
Grant Year
2003
Project Director
Yang, W.
Recipient Organization
ALABAMA A&M UNIVERSITY
4900 MERIDIAN STREET
NORMAL,AL 35762
Performing Department
FOOD SCIENCE & ANIMAL INDUSTRY
Non Technical Summary
A novel system will be developed for rapid weight approximation of hogs without placing stress on the animal or on the human operator. This system will consist of a video camera, a sensor to determine the distance to the animal, and a video display monitor. This will give producers a better estimate of shipping weights and reduce error costs.
Animal Health Component
(N/A)
Research Effort Categories
Basic
(N/A)
Applied
100%
Developmental
(N/A)
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
50335102020100%
Knowledge Area
503 - Quality Maintenance in Storing and Marketing Food Products;

Subject Of Investigation
3510 - Swine, live animal;

Field Of Science
2020 - Engineering;
Goals / Objectives
The objectives of this project are to: 1) increase the accuracy of the computer model based on image processing and neural networks for swine weight estimation, and 2) develop a portable, hand-held unit for pig weight approximation that are self-calibrating, fully automated and user-friendly.
Project Methods
The development of the weight approximation system will be accomplished in three stages. The first stage will be the experimental phase to gather data on the animals, which will be done by imaging a large cross-section of animals and recording their corresponding weights. The second stage will be the analysis of the collected data. A number of features of the animals will be extracted from the images and correlated to the weights. The third step will be to develop a portable imaging unit consisting of a video camera, a sensor to determine the distance to the animal, and a video display monitor. These will be mounted in a light, hand-held enclosure intended to be portable and easily used by a single operator. The sensor and camera will be connected to a small computer unit that will be carried in a backpack. The computer will be a regular Pentium PC and will also include a frame grabber card for digitizing the video signal. The entire system will be battery powered. The top features extracted will be used to train a neural network which will in turn be used to predict animal weights with the portable weight approximation system.

Progress 09/15/03 to 09/14/05

Outputs
The strength of correlation between the pig weight and measurable features was studied. It was found that the area and rear area gave the best correlation to weight (r=0.96). The correlation coefficients of the center width and flank width to weight were also high (0.94 - 0.95). Rear area was therefore chosen for developing a mathematical model to estimate hog weight. A protocol has been developed which allows the measurement of weight of a pig as it walks under a camera. This method was proven to be fast, relatively accurate and less stressful to both the animal and the operator. The conventional area extraction method has been improved using the Canny edge detection approach. The accuracy of the edge detection-based area extraction was analyzed. A lab-scale portable weight approximation system for pigs is being developed and tested.

Impacts
This project will enhance the abilitity to check the weight of an animal with out physically touching the animal.

Publications

  • Wang, Y., W. Yang, P. Winter and L.T. Walker. 2006. Non-Contact Sensing of Hog Weights by Machine Vision. Applied Engineering in Agriculture, 22(4): 577-582.
  • Wang, Y., W. Yang, P. Winter and L.T. Walker. 2006. Developments in machine vision based swine weighing. Abstract No. 078E-13, Institute of Food Technologists Annual Meeting, Orlando, FL, June 24-28, 2006.
  • Wang, Y., W. Yang, P. Winter and L.T. Walker. 2006. Walk-through weighing of hogs by machine vision and neural networks. Abstract No. 078E-14, Institute of Food Technologists Annual Meeting, Orlando, FL, June 24-28, 2006.
  • Wang, Y., W. Yang and P. Winter. 2006. Accuracy of projected area extraction in pig weighing by machine vision. Abstract No. 078E-15, Institute of Food Technologists Annual Meeting, Orlando, FL, June 24-28, 2006.
  • Yang, W. and P. Winter. 2004. Development of an image and neural network based technology for weight approximation of animals. Abstract No. 99B-25, Institute of Food Technologists Annual Meeting, Las Vegas, NV, July 12-16, 2004.


Progress 01/01/04 to 12/31/04

Outputs
The technique under development in this project is non-contact weighing of pigs. A pig imaging setup using a EDC 3000B digital camera was established. The images were processed using the Data Translation Global Lab Image/2, and the extracted features were analyzed using a Matlab Neural Network toolbox. Experiments are being conducted to refine the weight approximation model developed in the preliminary study. Its accuracy of prediction is expected to be improved in the next phase of experiments.

Impacts
This project will enhance the abilitity to check the weight of an animal with out physically touching the animal.

Publications

  • No publications reported this period