Corn Ear Ripeness Ratings Based On The Wavelet Analysis

Ripeness is one of the most important quality indicators during the course of corn ear processing. Today, the ratings of corn ear ripeness are only based on human testing under industry guidance without national standards. Such ratings method tends to be more subjective with results always varying between different testers, hence the accuracy can hardly be guaranteed; furthermore, human testings tend to be low-efficient and might even cause comtamination to the corn ear. Therefore, the computer visual technology is introduced, which supports less subjective and highly efficient quantitive testing, causing no damage to the core ear. Such technology has yet to be applied prevalently in the real world but is worth further studying.The wavelet is a new multi-demensional analytical procedure, analysing signals on various scales. This paper takes the wavelet method in analysing and extracting eigenvalue of the image of corn ear. It also adopts the Bayes stepwise discriminant analysis and the GRNN neural network respectively in building up models for computerised core ear rating purposes. The major research and findings are:1. Design the corn ear ratings system by using industrial cameras as the image acquisition tools and fluorescent as the illuminating system and all image acquisitions are undergoing in a sealed box in order to prevent natural light from affecting the quality of image acquisitions.2. Analyse the relevance between the features of color histogram and the ripeness of corn ear. Extract the color information of images under RGB and HSV models. Through single-dimensional wavelet decomposition, the maximum points'coordinates, mean value and standard deviation can be extracted. Through analysis of main components, dimensionality of the eigenvalue can be reduced. As such, we can abtain the main components of 3 features of color histogram. The accuracy of testing by applying the Bayes stepwise discriminant analysis to the 3 features of color histogram can be as high as 90.9%. The accuracy of testing by applying the GRNN neural network (with the smooth factor as 0.1) is 78.6%.3. Analyse the relevance between the color eigenvalue E and the ripeness of corn ear. Under the RGB and HSV models, decompose the acquired two-dimensional image to get 6 color components– R, G, B, H, S, V. Decompose it using two-dimensional wavelet analysis and extract color eigenvalue E of the 6 wavelet lower frequency parts which include most color information of the images. The accuracy of testing by applying the Bayes stepwise discriminant analysis is 88.2% under the RGB model and 87% under the HSV model. Considering all six color eigenvalue E of R, G, B, H, S, V, the accuracy is 88.5%. The accuracy of testing by applying the GRNN neural network is 71.0% (with the smooth factor as 0.5). 4. Analyse the relevance between the wavelet vein features and the ripeness of corn ear. After image grayness, veins and marginal information are mainly included in the high frequency parts due to wavelet decomposition. Decompose it using 3 dimensional wavelet analysis and the accuracy of testing by applying the Bayes stepwise discriminant analysis is 89.7%. The accuracy of testing by applying the GRNN neural network is 75.6% (with smooth factor as 0.5).5. Through wavelet analysis of color and vein of the corn ear images, compare the testings of Bayes stepwise discriminant analysis and GRNN neural network. The results indicate that colors and veins of the corn ear are the effective information in rating the ripeness of corn ear. In general, the accuracy by applying the Bayes stepwise discriminant analysis is higher than the GRNN neural network. Among all testings, the Bayes stepwise discriminant analysis applying to the color histogram obtains the highest accuracy of 90.9%.