The Visioneers » Milled Rice

We got 2nd during the 32nd R&D Review

whaldsz — Fri, 01 Jul 2011 21:24:19 +0000

During the 32nd In-house Research and Development Review last May10-11, 2011 held at Munoz City, Philippines. Our on-going research on Milled Rice Computer Vision garnered an award.

For more information, follow this link - http://www.philmech.gov.ph/?page=news&action=details&code01=NP11050002

Share now!

Caliper Software for Milled-Rice – Counts the number of grains, headrice and broken kernels

whaldsz — Sun, 14 Sep 2008 02:53:02 +0000

This is an application for counting milled rice grains, determining the count of headrice, broken, and brewers. Using these grade factors, the caliper software (as I call it) is able to estimate the total weight in terms of percentage. Weight estimation of rice grain is performed using linear regression and support vector machines (SVM). It uses AForge library from AForge.NET for various image processing task.

I did this software for Bureau of Post-harvest Research Extension (BPRE) and indirectly, to National Food Authority (NFA) for the purpose of quickly determining the grain size of milled rice.

The demo and initial version of the caliper software is available for download here:

caliper software.

To use the software, simply install the application, then open the sample image included in the installation (“bigas.bmp”), then that’s it! You can try it for other similar problems, like corn, barley, etc. Please give me feedback if problems arises. For any questions, you may email me at {vlad_crasher at yahoo point com}.

UPDATE: Yesterday (September 15, 2008), they told me that the test was successful! The classification result did matched with the manual methods performed by human inspector.

Share now!

Rice quality evaluation

whaldsz — Thu, 21 Aug 2008 14:33:33 +0000

Prior to the previous post, I also finished a paper entitled “Automatic milled rice quality evaluation” that’s about to appear in one international conference proceedings. The idea is to build a low-cost, efficient, and reliable alternative for the milled rice quality evaluation. As you know, some rice quality standards require the grading factors or defectives in terms weight (grams) , making this problem a very challenging and interesting one! Read on, some results are discussed in the succeeding paragraphs.

This is the block diagram of the milled rice quality evaluation system, comprised of 6 different stages.

Sample milled rice image is shown below, (a) acquired from scanner and (b) background-segmented image. It is important that unnecessary objects are removed prior to further image processing.

These are the segmented kernel images prior to features extraction. Different types of kernels are obviously visible, paddy, chalky, damaged, broken, good/sound kernels.

We then use the extracted features from the above kernel images to classify according to defective types. The result is then processed to determine the estimated weight. Finally, based on these estimates, we judge according to rice quality standards of course, whether the rice is premium grade, grade 1, grade 2, etc. The result in terms of weight estimates were quite exceptional and the classification, too! However, I’m still looking for a much better approach that gives better accuracy. I built my classifier using probabilistic neural network (PNN). And oh, I developed this system in C#.

Any better suggestion for improving my regression and classifier? I welcome your comments and suggestions….

Share now!

Weight estimation of milled rice

whaldsz — Wed, 20 Aug 2008 17:10:03 +0000

This article presents a method for weight estimation and classification of milled rice kernels using supervised learning algorithm. Shape descriptors are used as geometric features for determining the grade factors such as headrice, broken kernel, and brewer. Color histogram is extracted from milled rice image to obtain 24 color features in RGB and Cielab color spaces. A support vector machines (SVM) is adopted for addressing the classification and regression problem in milled rice quality evaluation. We built a support vector regression (SVR) model for estimating rice kernel weight and support vector classifier (SVC) for rice defectives.

Results showed that in real data, the performance of SVR is better than linear regression (LR) with a mean square error (MSE), mean absolute error (MAE) and correlation coefficient of 0.078, 0.21 and 99.4%, respectively. For classification of rice defectives using SVC, the accuracy is 98.9% outperforming the general regression neural network (GRNN) model.

Share now!