关键词: 檀香, 图像分割, 营养诊断, 氮素, 颜色系统

Abstract:

Objectve: In order to ensure the survival rate of sandalwood (Santalum album) and the quality of heartwood in later period, this paper proposed an image segmentation method of sandalwood and a soil available nitrogen nutrition method which was expected to provide a time-saving method to monitor the growth of sandalwood. Method: Converting an image from RGB to a HSI system, then Otsu method to S and I channel was applied in this study. Combining the advantages of the above channels and image filtering method as well as morphological operation method, the sandalwood leaves were segmented from complex background. Using RGB, HSI and Lab systems, the color mean values of leaf images were calculated respectively, and soil available nitrogen content prediction models were built under different fertilization levels and under all levels. For each color system, the mean value of single channel of sandalwood leaves was taken as an independent variable, and the available nitrogen content of each sandalwood tree was taken as a dependent variable to establish a quadratic polynomial of three variables. By calculating the model validation index of fitting data and validation data, the best model was determined. Result: 1) In sandalwood segmentation method under complex background, the S channel could divide the green plants into a whole part, while the I channel could distinguish sandalwood leaves from the other plant leaves. The combination of the two channels could successfully remove most of the background. Combining 7×7 median filter, morphological operation and super G factor, the foreground was extracted more accurately. The pixel number error of this algorithm was within 5%, and the average error of each color channel was controlled within 2%, which showed that the segmentation algorithm was feasible. 2) When building the prediction model of soil available nitrogen content, we compared the prediction result of different color systems. It was found that the Lab system could reflect soil available nitrogen content more accurately under different nitrogen application levels. So dumb variable method was used to build the model, and the prediction model of soil available nitrogen content based on dumb variable was also obtained. At the same time, considering the unknown level of nitrogen application in some forest farms due to neglect of management, this study established a prediction equation at the full level. The result showed that the Lab system was still the best one. The parameter test of the two models presented significant result, indicating that the effects were relatively ideal. Conclusion: The sandalwood leaves could be extracted accurately by combining the characteristics of S channel and I channel in HSI color system under Otsu method, and using median filter, morphological operation and super G factor for post-processing also guaranteed the accuracy. Based on the image color parameters obtained from complex background segmentation, the diagnosis of nitrogen nutrition sandalwood was carried out in this paper. We discovered that the Lab was the best color system regardless of whether nitrogen level was divided or not, and its quadratic polynomial model presented a good prediction ability. The method proposed in this paper could quickly obtain soil nitrogen nutrient content for managers and might ensure the healthy growth of sandalwood.

Key words: sandalwood(Santalum album), image segmentation, nutrition diagnosis, nitrogen, color system