| Summary: | Facilitating non-contact measurement, a computer-imaging system was devised and evaluated to predict the chlorophyll content in potato leaves. A charge-coupled device (CCD) camera paired with two optical filters and light chamber was used to acquire green (550 ± 40 nm) and red band (700 ± 40 nm) images from the same leaf. Potato leaves from 15 plants differing in coloration (green to yellow) and age were selected for this study. Histogram based image features, such as mean and variances of green and red band images, were extracted from the histogram. Regression analyses demonstrated that the variations in SPAD meter reading could be explained by the mean gray and variances of gray scale values. The fitted least square models based on the mean gray scale levels were inversely related to the chlorophyll content of the potato leaf with a R2 of 0.87 using a green band image and with an R2 of 0.79 using a red band image. With the extracted four image features, the developed multiple linear regression model predicted the chlorophyll content with a high R2 of 0.88). The multiple regression model (using all features) provided an average prediction accuracy of 85.08% and a maximum accuracy of 99.8%. The prediction model using only mean gray value of red band showed an average accuracy of 81.6% with a maximum accuracy of 99.14%. Keywords: Computer imaging, Chlorophyll, SPAD meter, Regression, Prediction accuracy
|
|---|
