A.V. De Grano and J.P. PABICO. 2007. Transactions (ISSN 0115-8848) of the National Academy of Science and Technology of the Philippines 29(1):131-132. Best Scientific Poster in the Engineering Sciences and Technology Division.

Abstract

Color in tomato (Lycopersicon esculentum) is the most important external characteristic to assess ripeness and postharvest life, and is a major factor in the consumer’s purchase decision. The degree of ripening is usually estimated visually by human graders comparing the tomato color to a chart that classify fresh tomatoes into six maturity stages according to the USDA standard classification: Green, Breakers, Turning, Pink, Light Red, and Red. This manual practice of tomato maturity classification often results into errors due to human subjectivitiy, visual stress, and tiredness. We developed a color image analysis procedure and a neural network model to automate the classification of tomato maturity. We captured using a computer-connected digital camera 6,000 color images of locally grown and harvested tomatoes equally representing the six maturity stages (1,000 each). The average classification by five expert graders from a local commercial farm was used as the maturity classification of each tomato. Using the red, green and blue (RGB) spectral values of the captured images as inputs, we trained a neural network-based tomato maturity classifier to indicate the degree of maturity within each stage and to provide a continuous index over the complete maturity range. We trained a 3-layer neural network via the feed-forward, back propagation training algorithm using 70% of the captured images as the training set (4,200 images) and 10% as the test set (600 images), equally representing each maturity stage. The test set was used during training to avoid model over-fitting. Validation results agreed with manual grading in 97% of the remaining tomatoes (1,200 images), while the remaining 3% were classified wrongly but within one maturity stage difference. With this result, an automatic vision system for tomato grading could be a potent alternative to manual grading.

Keywords: tomato grading, neural network, image processing, feed-forward, back-propagation

Alona V. De Grano is an undergraduate alumna of ICS.

Submitted 15 January 2007, revised 30 March 2007, and presented 11-12 July 2007 as a Scientific Poster to the 29th Annual Scientific Meeting of the National Academy of Science and Technology, The Manila Hotel, Philippines. Awarded 12 July 2007 as Best Poster in the Engineering Sciences and Technology Division of the National Academy of Science and Technology of the Philippines.