| Summary: | 碩士 === 國立屏東科技大學 === 機械工程系 === 93 === MS medium is commonly used as a basic nutrition for plant tissue culture. To users, the contamination problem of MS medium causes significant loss in economy and reputation. This study applied multispectral image and Principal Component Analysis (PCA) to identify the biological contaminants in MS media with and without carbon added from normal. The image of each specimen was taken at spectrums of 300-400, 480-560, 560-620, 400-700, 400-770, 850-870, and 850-1050 nm from day 2 to day 7 consecutively after the samples were prepared. The features were extracted from the images by three PCA methods, i.e. single band multi-samples, multi-bands single sample and multi-bands multi-samples, respectively. The treatment of Smoothed Edge Eigenspace (SEE) was also tested to explore its effects on classification accuracy. Two classification methods named Bayesian classifier and Mahalanobis distance were applied to classify the normal and contaminated media. Results indicate that Mahalanobis distance with the SEE treatment combining with spectrums of 480-560, 560-620, and 850-870nm reach the classification rate of 88% for day 3 images. For carbon added media, the bands combination of 400-770 and 300-400 nm has the classification rate of 83% on day 2. In single band multi-samples method with Bayesian classifier, the result showed that 480-560nm spectrum can reach classification rate of 92% for day 5 samples.
|
|---|
