Senior Researcher Korea Institute of Science and Technology
Disclosure: Disclosure information not submitted.
Accurately and quickly calculating the leaf area index of cultivated crops is essential for monitoring the growth environment and physiology of corps. Recently, methods for measuring the leaf area of crops using imaging devices such as cameras and scanners, in conjunction with digital image processing and machine learning algorithm, have been steadily developed. However, there are several limitations in applying this method directly at the greenhouse site: overlapping of the digital images with adjacent crops and the inability to measure images on the back of the crops because the crops are densely grown. In this study, , in order to measure and calculate leaf area of the tomato canopy by in suit, non-destructive imaging method in the greenhouse, several new measurement techniques are introduced. First, using a RGB-depth camera with a wide field of view angle lens, the entire canopy image was combined by scanning at regular intervals vertically for a crop at close range. The shorter the image acquiring interval, the better to distinguish between the ROI and the surrounding area due to the difference in viewing angle. To accurately count only the number of leaf branches extending from a tomato crop to be measured, the difference between leaf branches extending from neighboring crops, leaf nodes covered by one's own leaves, and number of nodes by tomato flowers or fruits was processed by deep neural network method. The leaf area correction of the branches extending from one node was carried out using the pixel-by-pixel distance information in the captured depth information of the image. The total leaf area for a crop was calculated using the number of nodes and the calculated leaf area for a crop. In order to confirm the accuracy of this method, the same crops were measured in a destructive way and compared.