Visualization of Water Content Distribution in Potato Leaves Based on Hyperspectral Image
Sun Hong,Liu Ning,Wu Li,Zheng Tao,Li Min-zan,Wu Jing-zhu
DOI: https://doi.org/10.3964/j.issn.1000-0593(2019)03-0910-07
2019-01-01
Spectroscopy and spectral analysis
Abstract:In order to quickly detect the water content of potato leaves and explore the change of leaf water content under drought stress, the hyperspectral imaging technology was utilized to detect and visualize the moisture content of potato leaves in this paper. 71 leaves were collected and the water gradient of the leaves was controlled by a drying method. A total of 355 samples were obtained. The hyperspectral sorting instrument was used to collect potato leaves spectral and image data of 862. 9 similar to 4 704. 2 rim (256 wavelengths). The water content was measured by weighing method. According to a certain proportion, Sample Set Partitioning Based on Joint X-Y Distance (SPXY) algorithm was used to divide the sample into a model set and a validation set. For the calibration set, the feature wavelengths were extracted by using Coefficient Analysis (CA) and Random Frog (RF) algorithms respectively, and Partial Least Squares Regression (PLSR) models were established respectively. The calibration set and validation set determines coefficient R-2 and the RMSE (Root Mean Square Error) were used as the evaluation index. The gray image of the potato leaves water content was calculated using the results of the detection model. The visualization analysis of potato leaves water content was realized based on the pseudo color image transformation and segmentation. The average reflectance of each sample leaf was calculated by ENVI software, obtaining a total of 355 sample's spectral data. According to the proportion of 2, 1, the total samples were divided into calibration set (240 samples) and validation set (115 samples) by SPXY algorithm. Spectral characteristics of the collected data were analyzed. Two algorithms, CA and RF, were used to select 15 characteristic wavelengths, respectively. Based on the CA, the selected 15 wavelengths with the correlation coefficient higher than 0. 96 were 1 406. 82, 1 410. 12, 1 403. 62, 1 413. 32, 1 416. 62, 1 419. 82, 1 400. 32, 1 423. 12, 1 426. 32, 1 429. 62, 1 432. 82, and 1 441. 12, 1 493. 32, 1 442. 52 and 1 445.8 nm. Based on the RF algorithm, the 15 feature wavelengths that the selected probability higher than 0. 3 were 1 071. 62, 1 041. 12, 1 222. 52, 1 465. 22, 1 397. 02, 1 449. 02, 1 034. 32, 1 523. 22, 976. 42, 1 172. 52, 979. 82, 1 165. 82, 1 037. 72, 1 426. 32 and 869.8 nm. The PLSR model was established using the characteristic wavelengths filtered by the CA and RF algorithms, marked RF-PLSR and RF-PLSR model respectively. The water content of potato leaves was analyzed visually using the more precise model. First, the each pixel water content of the potato leaf image was calculated to obtain a gray image. Then, the gray image was pseudo-color transformed to draw a visual color image of the leaf water content. In order to reflect the change of potato leaves water content in the drying process, HSV model was used to segment the pseudo-color image of sample leaf. The segmentation image results, showing the proportion of leaf area in a certain water content interval, were obtained. The results showed that the 15 wavelengths selected by the CA algorithm were in the range of 1 400. 3 to 1 450. 0. The calibration accuracy of CA-PLSR was 0. 975 5, and the RMSEC (Root Mean Square Error of Calibration) was 2. 81%, and the validation accuracy was 0. 933 2, and the RMSEV (Root Mean Square Error of Validation) was 2. 31%. The range of characteristic wavelengths selected by the RF algorithm, with local "peak valley" characteristic, was wider than that of the CA algorithm. The calibration accuracy of RF-PLSR model was 0. 983 2, and the RMSEC was 2. 32%, and the validation accuracy was 0. 947 1, and the RMSEV was 2. 15%. The RF-PLS model is selected to calculate the water content of each pixel in the potato leaf images. According to the pseudo-color image, it could be seen intuitively that the water content gradually decreased with the drying time increasing. From the perspective of leaf tissue structure, with the strengthening of water stress, the leaves began to lose water from the edge and gradually spread to the middle, in which the water content of leaf stems and veins was higher than that of other parts. The pseudo-color image was segmented by HSV model based on the color difference of the image. The proportion of pixels with water content greater than 90%, 80%, and 70% in the leaf pseudo-color image to the entire leaf image. Using the hyperspectral imaging technology can realize the water content detection and distribution visualization of potato leaves, which provides a new theoretical basis for the potato growth analysis and potato leaf water content analysis.