Multi-spectral imaging technique based on texture analysis and machine learning was proposed to discriminate alien
invasive weeds with similar outline but different categories. The objectives of this study were to investigate the
feasibility of using Multi-spectral imaging, especially the near-infrared (NIR) channel (800 nm±10 nm) to find the
weeds' fingerprints, and validate the performance with specific eigenvalues by co-occurrence matrix. Veronica polita
Pries, Veronica persica Poir, longtube ground ivy, Laminum amplexicaule Linn. were selected in this study, which
perform different effect in field, and are alien invasive species in China. 307 weed leaves' images were randomly
selected for the calibration set, while the remaining 207 samples for the prediction set. All images were pretreated by
Wallis filter to adjust the noise by uneven lighting. Gray level co-occurrence matrix was applied to extract the texture
character, which shows density, randomness correlation, contrast and homogeneity of texture with different algorithms.
Three channels (green channel by 550 nm±10 nm, red channel by 650 nm±10 nm and NIR channel by 800 nm±10 nm)
were respectively calculated to get the eigenvalues.Least-squares support vector machines (LS-SVM) was applied to
discriminate the categories of weeds by the eigenvalues from co-occurrence matrix. Finally, recognition ratio of 83.35%
by NIR channel was obtained, better than the results by green channel (76.67%) and red channel (69.46%). The
prediction results of 81.35% indicated that the selected eigenvalues reflected the main characteristics of weeds'
fingerprint based on multi-spectral (especially by NIR channel) and LS-SVM model.
Tea categories classification is an importance task for quality inspection. And traditional way for doing this by human is time-consuming, requirement of too much manual labor. This study proposed a method for discriminating green tea categories based on multi-spectral images technique. Four tea categories were selected for this study, and total of 243 multi-spectral images were collected using a common-aperture multi-spectral charged coupled device camera with three channels (550, 660 and 800 nm). A compound image which has the clearest outline of samples was process by combination of the three monochrome images (550, 660 and 800 nm). After image preprocessing, 18 morphometry parameters were obtained for each samples. The 18 parameters used including area, perimeter, centroid and eccentricity et al. To better understanding these parameters, principal component analysis was conducted on them, and score plot of the first three independent components was obtained. The first three components accounted for 99.02% of the variation of original 18 parameters. It can be found that the four tea categories were distributed in dense clusters respectively in score plot. But the boundaries among them were not clear, so a further discrimination must be developed. Three algorithms including support vector machines, artificial neural network and linear discriminant analysis were adopted for developed classification models based on the optimized 9 features. Wonderful result was obtained by support vector machines model with accuracy of 93.75% for prediction unknown samples in testing set. It can be concluded that it is an effective method to classification tea categories based on computer vision, and support vector machines is very specialized for development of classification model.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.