GRADES ARE IMPORTANT
In most countries, including the Philippines, many agricultural products are still graded using a standard criterion. In tomatoes, color is used to determine ripeness; in eggs, the shells are inspected for defects.
The color of tomato and the shell quality of eggs affect consumers’ purchasing habits, hence, dictate the price of the product. Accurate grading is therefore critical for enterprises.
The problem with visual inspection of tomatoes and eggs, however, is that of accuracy over time. “Because manual grading is labor-intensive, workers will naturally grow tired and bored— thus increasing the chances for errors, especially after they have worked for a few hours without break,” explained Prof. Pabico.
SOLUTION: AUTOMATE THE GRADING
Although several ways of automated grading of agricultural products have been developed abroad, the systems are costly and would only work in controlled farm environments. The Philippines needed something for its small farms.
Prof. Pabico’s model provides an affordable alternative. He came up with it by setting up machine vision systems (MVS) – two computers equipped each with a web camera. The first MVS will be used for grading tomatoes while the second, for eggs.
The camera serves as the “eye” of the MVS that senses and captures images of the tomatoes or eggs, while the computer runs the “brain” of the MVS, which determines the grade of the produce.
CREATING THE BRAIN
Prof. Pabico said that building the MVSs is the easy part. It is creating and optimizing the brains for each system that took time.
“We took the first MVS set to a commercial tomato farm in Tagaytay City and took 6,000 pictures of freshly harvested tomatoes using the web camera. The other unit we used to capture 750 images of harvested eggs from a backyard poultry raiser in Sariaya, Quezon,” Prof. Pabico recounted.
The images of the tomatoes and eggs were then prepared for ‘feeding’ into the respective brains—an artificial neural network (ANN). The process involved extracting the image of the product from its background using what is called an edge algorithm, extracting the red/green/blue spectral patterns of the product, and normalization of the patterns. The normalized patterns were used to ‘train’ the MVSs to grade the tomatoes and eggs.
After running tests, the computers determined whether the tomatoes where in the Green, Breakers, Turning, Pink, Light Red, and Red stages and whether the eggs are acceptable or rejects.
BRAIN WINS AGAINST BRAWN
After measuring performances of both human graders and the ANN in classifying eggs, Prof. Pabico found that the ANN posted a grading performance of 76%; humans posted 73%. ANN also did the job of classifying tomatoes better, recording an accuracy of 97% against the humans’ 93%.
In addition, the average accuracy of the human tomato and egg graders declined over an eight-hour work shift. Human accuracy was found to go up again after 15-minute snack breaks and after lunch, but the performance increase was not enough to bring the accuracy back to original levels.
The ANNs developed in this research may be used as a potent classifier for MVSs for tomato maturity classification and egg grading,” said Prof. Pabico.
ANOTHER BRIGHT IDEA
“Later on we hope to look into the rate of grading of both humans and the MVS by hooking up the MVS to a conveyor belt. We can then look for ways to speed up image acquisition and processing, as well as the analysis by the ANN and its data output on screen,” he added.
“With current advances in ANN research, it may be soon possible to grade agricultural produce using a group of ANNs that will analyze objects simultaneously. This would increase accuracy even more,” he concluded.
21.08.2017 | Albert-Ludwigs-Universität Freiburg im Breisgau
AI implications: Engineer's model lays groundwork for machine-learning device
18.08.2017 | Washington University in St. Louis
16.08.2017 | Event News
04.08.2017 | Event News
26.07.2017 | Event News
21.08.2017 | Medical Engineering
21.08.2017 | Materials Sciences
21.08.2017 | Life Sciences