ISHS Acta Horticulturae 440: International Symposium on Plant Production in Closed Ecosystems
TEXTURAL FEATURES-BASED NEURAL NETWORK MODEL VS. SOIL COVERAGE MODEL FOR BETTER ADAPTABILITY IN NON-DESTRUCTIVE GROWTH MEASUREMENT OF PLUG SEEDLINGS COMMUNITY |
| Authors: | T. Suzuki, H. Murase, Y. Nishiura, H. Takigawa, N. Honami |
| Keywords: | image processing, neural network, plug seedlings community, soil coverage, textural features, top fresh weight |
| Abstract:
The total top fresh weight of cabbage plug seedlings must be estimated properly to identify their growth stage based on obtained digital image of their community. To relate the digital image and the top fresh weight, the following two models were tested. The first was a textural features-based neural network model and the second was a soil coverage model. The three different soil coverage models were used to calculate the relative soil coverage of the plug seedlings community based on the image processed at the respective threshold value of the chrominance level of G component (Cg). The Cg was calculated from the RGB signals for color images of the community of plug seedlings. The textural features-based neural network model was also tested using the same data as that used for the soil coverage model determination. Angular second moment, contrast and inverse difference moment textural features were selected. A Kalman neuron training algorithm was used to model the relationship between the top fresh weight of the plug seedlings community and the selected textural features. The performance of each model was determined by comparing the measured data with estimated data. There were only marginal differences in their performance. Each model estimated top fresh weight at acceptable accuracy from the given image data. | |
Download Adobe Acrobat Reader (free software to read PDF files) | |
