Determination of moldy corn kernels from bulk samples using image analysis
2005
Bulaong, M.C.
A computer vision system for determining weight percentage of moldy kernels (WPMK of corn (Zea mays L.) in bulk samples, consisting of a flatbed scanner and a desktop computer equipped with an image processing and artificial neural network, (ANN) software was developed. Images of corn samples collected from four major corn producing areas in the Philippines were acquired, processed and analyzed using the system. Color and texture features were extracted from each image using an open source image processing software (Image J) and used as inputs for training, validation and testing of models based on canonical discriminant analysis (CDA) and ANN model. The ANN performed consistently better than CDA in classifying corn grains into different WPMK levels. The ANN model based on color features extracted from red, green, and blue (RGB) histogram and texture features extracted from hue, saturation, and brightness (HSB) color space obtained the highest correlation coefficient of 0.95 between predicted and actual percentage moldy kernels. Total time for image acquisition and processing is approximately 30 minutes using 10 scanning intervals. This is faster than manual mold analysis which takes about one hour. The study proved that a flatbed scanner can be used as a low-cost but rugged alternative to digital camera for computer vision system for fast, accurate and objective determination of WPMK from images of bulk corn samples with potential application for other grains.
显示更多 [+] 显示较少 [-]