论文部分内容阅读
Abstract In view of the shortage of using traditional methods to monitor chlorophyll content, hyperspectral technology was used to estimate the chlorophyll content of apple leaves rapidly, accurately and non??destructively. Based on the data of hyperspectral reflectivity and SPAD value of normal apple leaves and the leaves under the stress of red spiders collected from the Wanjishan base in Tai??an, the correlations of SPAD value with the original spectral reflectivity of apple leaves and its first derivative and between SPAD value and high spectral value were analyzed to select sensitive bands, and the estimation models of chlorophyll content in apple leaves based on hyperspectral reflectivity were established. The sensitive bands of chlorophyll content in normal apple leaves were 513-539, 564-585, 694, 699 and 720 nm, and the best estimation model of chlorophyll content was SPAD=152.450-1 884.851 R377. The sensitive bands of chlorophyll content in the leaves under the stress of red spiders were 961, 972 and 720 nm, and the best estimation model of chlorophyll content was SPAD=49.371-46 428.473 R??972.
Key words Hyperspectral data; Apple; Chlorophyll; Spectral features; Correlation
Chlorophyll content, an important biochemical parameter in process of plant growth, can reflect the photosynthetic capacity, developmental stage and nutritional status of vegetation[1]. At present, the methods commonly used to monitor chlorophyll content are spectrophotometry and SPAD??502 chlorophyll meter. Traditional spectrophotometry is time??consuming and laborious, and it is difficult to meet the requirements of real??time, fast, non??destructive and large??area monitoring of precision agriculture. The hand??held SPAD??502 chlorophyll meter produced by Japan??s Minolta Camera can only monitor leaves point by point, and needs to measure the average of chlorophyll content in multiple plants as measurement results, so the workload is large[2]. Hyperspectral remote sensing is a rapid and non??destructive monitoring technique that monitors the nutritional status of crops during the growing season without destroying the tissues of plants[3].
Domestic scholars?? research on hyperspectral data of apple focuses on hyperspectral estimation of crown LAI[4] and SPAD[5], estimation of chlorophyll content in apple leaves based on RGB model[6], hyperspectral estimation of chlorophyll content[7], extraction of diseased areas of apple leaves[8], and so on. There are few reports on the estimation of chlorophyll content in normal and diseased leaves of apple trees in Shandong. In this study, based on the data of hyperspectral reflectivity and chlorophyll content of apple leaves, the correlation between the hyperspectral reflectivity and chlorophyll content was analyzed, and quantitative relationship models between the chlorophyll content and spectral characteristic parameters of apple leaves were established to provide theoretical basis and technical support for the growth monitoring of apple by using hyperspectral remote sensing technology. Materials and Methods
Experimental design
On June 8, 2017, the leaves of apple trees used in this experiment were collected from the Wanjishan base in Tai??an, placed in freshness protection bags, and brought back to a laboratory under conditions of low temperature and no damage.
The damage of leaves infected with red spiders was divided into primary, intermediate and advanced leaves according to the number of red spiders on the leaves. Normal leaves (defined as level 0) were collected as the control group. Ten leaves were picked from each level of damage, and a total of 40 apple leaves were collected.
Spectrometry
The hyperspectral data of leaves of apple trees were obtained by using the SOC710VP visible??near??infrared hyperspectral imaging spectrometer produced by Surface Optics Corporation of the United States in a laboratory where light conditions (light was provided by tungsten lamps) could be controlled. The spectral range was 350-1 050 nm, and the spectral resolution was 4.687 5 nm.
SPAD values of apple leaves
After apple leaves were picked, the SPAD values of apple leaves were measured immediately using a SPAD??502 chlorophyll meter, and each of the leaves was marked to correspond to the measured hyperspectral data.
Results and Analysis
Extraction and analysis of spectral characteristics of leaves under stress of red spiders
Under the stress of red spiders (the leaves were attached with red spiders), there was no obvious difference between the leaves under different leaves of stress (primary, intermediate and advanced stress) in terms of spectral reflectivity, but the difference between these diseased leaves and normal leaves was obvious at 420-700 and 750-1 050 nm (Fig. 1).
In the four states (level 0, primary, intermediate and advanced level), there was a consistent changing trend of hyperspectral reflectivity of apple leaves. That is, there were reflection peaks at 380 and 550 nm, an absorption valley at 680 nm, red edges at 680-780 nm, and a near??infrared reflection platform after 780 nm. In the near??infrared band (780-1 050 nm), the reflectivity of normal leaves was the highest, while that of the primary damaged leaves was the lowest, and that of the advanced and intermediate damaged leaves was moderate. This indicates that the internal cell structure of normal leaves was normal, and reflection was formed many times, so the reflectivity of normal leaves was the highest. The internal cells of leaves under the stress of red spiders were destroyed, so the reflectivity of the diseased leaves was lower than that of normal leaves. As the degree of stress deepened, the reflectivity of the red spiders attached to the leaves gradually increased with the increase of band, so the reflectivity of the advanced leaves was gradually higher than that of the intermediate leaves in the near??infrared band (850-1 050 nm).
There were big differences between normal leaves and the leaves under the stress of red spiders at 420-700 and 750-1 050 nm, which can be used as the recognition bands for judging whether the leaves suffer from the stress of red spiders.
The spectral reflectivity of red spiders also showed certain vegetation characteristics. In the near??infrared band (780-1 050 nm), the reflection of red spiders led to the increase of reflectivity of leaves, which can also provide reference for the classification of red spider disease index.
In the range of 400-780 nm, there was no obvious difference between the apple leaves under the stress of red spiders in terms of spectral reflectivity.
Extraction of red edge position
The red edge is the point where the reflectivity of green plants increases fastest at 680-750 nm and the inflection point of the first derivative in this interval, which is caused by the strong absorption of plants in the red light band and the strong reflection in the near??infrared band.
Seen from Fig. 2, the changing trends of the first derivative of normal leaves and the advanced leaves were similar, and there was an obvious trough and peak at 400 and 720 nm respectively. Red edge position was extracted by using the band where the maximum of the first derivative was located. Red edge position was at 720 nm.
Correlation analysis of the spectral reflectivity and its first derivative of apple leaves under the stress of red spiders with chlorophyll content
In the early period of apply growth, there was no big difference between the primary, intermediate and advanced leaves damaged by red spiders in terms of reflectivity, so the three kinds of leaves can be called the leaves under the stress of red spiders. Here the advanced leaves damaged by red spiders were analyzed.
The correlation between the original spectral reflectivity and chlorophyll content of normal leaves was analyzed, and SPSS statistical analysis was conducted. As shown in Fig. 3, there was a negative correlation between the original spectral reflectivity and chlorophyll content of normal leaves. The negative correlation was significant (P The correlation between the original spectral reflectivity and chlorophyll content of the advanced leaves was also analyzed, and SPSS statistical analysis was conducted. Seen from Fig. 3, the correlation between the original spectral reflectivity and chlorophyll content of the advanced leaves was not significant, and the correlation coefficient was small, but there was also an obvious trough and peak.
The correlation between the first derivative of original spectral reflectivity and chlorophyll content of normal leaves was analyzed, and SPSS statistical analysis was conducted. According to Fig. 4, there was an extremely significant negative correlation between them at 513-539, 564-585, 694 (Pearson correlation coefficient was -0.822**) and 699 nm (Pearson correlation coefficient was -0.877**).
There was a significant correlation between them at 372 (Pearson correlation coefficient was -0.672*), 498 (Pearson correlation coefficient was -0.694*), 508 (Pearson correlation coefficient was -0.739*), 544 (Pearson correlation coefficient was -0.739*), 549 (Pearson correlation coefficient was -0.749*), 559 (Pearson correlation coefficient was -0.758*), 590 (Pearson correlation coefficient was 0.736*), 611 (Pearson correlation coefficient was 0.759*), 642 (Pearson correlation coefficient was 0.716*), 705 (Pearson correlation coefficient was -0.759*), 816 (Pearson correlation coefficient was 0.749*), 843 (Pearson correlation coefficient was -0.726*), 945 (Pearson correlation coefficient was 0.674*) and 951 nm (Pearson correlation coefficient was 0.771*).
The correlation between the first derivative of original spectral reflectivity and chlorophyll content of the advanced leaves was also analyzed, and SPSS statistical analysis was conducted. Seen from Fig. 4, the correlation was significantly positive at 961 nm, and Pearson correlation coefficient was 0.632*. It was significantly negative at 972 nm, and Pearson correlation coefficient was -0.723*.
Seen from the above analysis, for the normal leaves of apple, the original spectral reflectivity at 377 nm, the first derivative of original spectral reflectivity at 513-539, 564-585, 694 and 699 nm, and the original spectral reflectivity at 720 nm (red edge position) were chosen as sensitive variables.
For the advanced leaves of apple, the first derivative of original spectral reflectivity at 961 and 972 nm and the original spectral reflectivity at 720 nm (red edge position) were chosen as sensitive variables. Establishment of models based on sensitive bands
Using SPSS software, the variables in the sensitive bands were transformed to establish various regression models, and the models with higher fitness in each transformation were found (Table 1). According to Table 1, the R2 value of the linear function estimation model that was determined by using the original spectral reflectivity at 377 nm as the variable was the largest, up to 0.427*. The R2 value of the logarithmic function estimation model that was determined by using the original spectral reflectivity at 377 nm as the variable was the second largest, up to 0.425*. The two estimation models were used for further model verification.
Seen from Table 2, the R2 value of the linear function estimation model that was determined by using the first derivative of original spectral reflectivity at 972 nm as the variable was the largest, up to 0.523*. The R2 value of the linear function estimation model that was determined by using the first derivative of original spectral reflectivity at 961 nm as the variable was the second largest, up to 0.400*. The two estimation models were used for further model verification.
Model verification
To test the accuracy and reliability of the estimation models, the test data of apple leaves measured in the same group of experiments were randomly selected to test and verify the estimation models of the selected SPAD values of apple leaves, and the models with high test accuracy were screened out (Table 3 and Table 4).
Conclusions and Discussion
Influenced by natural factors such as season, soil and climate and human factors such as fertilization, cultivation techniques and management, the hyperspectral information of apple leaves will change differently. In this study, the chlorophyll content of apple leaves in Shandong was monitored, and the models were verified based on the data of samples in the same region, which enhanced the credibility and adaptability of the monitoring models. However, whether these models apply to the detection of chlorophyll content in apple leaves in different regions and growth periods needs further exploration.
In this study, the relationship between chlorophyll content and hyperspectral characteristic parameters was used to establish the estimation models of chlorophyll content in apple leaves. After the accuracy test, the best estimation model for the chlorophyll content of normal apple leaves in Shandong was determined as follows: SPAD=152.450-1 884.851 R377. The best estimation model for the chlorophyll content of the advanced apple leaves was determined as follows: SPAD=49.371-46 428.473 R??972. These models provide a method and reference for the estimation of chlorophyll content in apple leaves, which has certain guiding significance and reference value for the precision fertilization and rapid and non??destructive growth monitoring of apple. References
[1] JIANG JB, CHEN YH, HUANG WJ. Using hyperspectral remote sensing to estimate canopy chlorophyll density of wheat under yellow rust stress[J]. Spectroscopy and Spectral Analysis, 2010,30(8):2243-2247. (in Chinese).
[2] WANG JH, HUANG WJ, LAO CL, et al. Inversion of winter wheat foliage vertical distribution based on canopy reflected spectrum by partial least squares regression method[J]. Spectroscopy and Spectral Analysis, 2007,27(2):1319-1322. (in Chinese).
[3] WANG KR, PAN WC, LI SK, et al. Monitoring models of the plant nitrogen content based on cotton canopy hyperspectral reflectance[J]. Spectroscopy and Spectral Analysis, 2011,31(7):1868-1872. (in Chinese).
[4] HAN ZY, ZHU XC, FANG XY, et al. Hyperspectral estimation of apple tree canopy LAI based on SVM and RF regression[J]. Spectroscopy and Spectral Analysis, 2016,36(3):800-805. (in Chinese).
[5] HAN ZY, ZHU XC, WANG L, et al. Hyperspectral evaluation of SPAD value of apple tree canopy based on continuum??removed method[J]. Laser & Optoelectronics Progress,2016,53(2):214-223. (in Chinese).
[6] CHENG LZ, ZHU XC, GAO LL, et al. Estimation of chlorophyll content in apple leaves based on RGB model using digital camera[J]. Acta Horticulturae Sinica,2017,44(2):381-390. (in Chinese).
[7] CHENG LZ, ZHU XC, GAO LL, et al. Hyperspectral estimation of phosphorus content for apple leaves based on the random forest model[J]. Journal of Fruit Science,2016,33(10):1216-1229. (in Chinese).
[8] HU RM,WEI M, JING X, et al. Research for extracting method of apple leaf ill spots based on hyperspectral image[J]. Journal of Northwest A & F University (Natural Science Edition),2012,40(8):95-99. (in Chinese).
Key words Hyperspectral data; Apple; Chlorophyll; Spectral features; Correlation
Chlorophyll content, an important biochemical parameter in process of plant growth, can reflect the photosynthetic capacity, developmental stage and nutritional status of vegetation[1]. At present, the methods commonly used to monitor chlorophyll content are spectrophotometry and SPAD??502 chlorophyll meter. Traditional spectrophotometry is time??consuming and laborious, and it is difficult to meet the requirements of real??time, fast, non??destructive and large??area monitoring of precision agriculture. The hand??held SPAD??502 chlorophyll meter produced by Japan??s Minolta Camera can only monitor leaves point by point, and needs to measure the average of chlorophyll content in multiple plants as measurement results, so the workload is large[2]. Hyperspectral remote sensing is a rapid and non??destructive monitoring technique that monitors the nutritional status of crops during the growing season without destroying the tissues of plants[3].
Domestic scholars?? research on hyperspectral data of apple focuses on hyperspectral estimation of crown LAI[4] and SPAD[5], estimation of chlorophyll content in apple leaves based on RGB model[6], hyperspectral estimation of chlorophyll content[7], extraction of diseased areas of apple leaves[8], and so on. There are few reports on the estimation of chlorophyll content in normal and diseased leaves of apple trees in Shandong. In this study, based on the data of hyperspectral reflectivity and chlorophyll content of apple leaves, the correlation between the hyperspectral reflectivity and chlorophyll content was analyzed, and quantitative relationship models between the chlorophyll content and spectral characteristic parameters of apple leaves were established to provide theoretical basis and technical support for the growth monitoring of apple by using hyperspectral remote sensing technology. Materials and Methods
Experimental design
On June 8, 2017, the leaves of apple trees used in this experiment were collected from the Wanjishan base in Tai??an, placed in freshness protection bags, and brought back to a laboratory under conditions of low temperature and no damage.
The damage of leaves infected with red spiders was divided into primary, intermediate and advanced leaves according to the number of red spiders on the leaves. Normal leaves (defined as level 0) were collected as the control group. Ten leaves were picked from each level of damage, and a total of 40 apple leaves were collected.
Spectrometry
The hyperspectral data of leaves of apple trees were obtained by using the SOC710VP visible??near??infrared hyperspectral imaging spectrometer produced by Surface Optics Corporation of the United States in a laboratory where light conditions (light was provided by tungsten lamps) could be controlled. The spectral range was 350-1 050 nm, and the spectral resolution was 4.687 5 nm.
SPAD values of apple leaves
After apple leaves were picked, the SPAD values of apple leaves were measured immediately using a SPAD??502 chlorophyll meter, and each of the leaves was marked to correspond to the measured hyperspectral data.
Results and Analysis
Extraction and analysis of spectral characteristics of leaves under stress of red spiders
Under the stress of red spiders (the leaves were attached with red spiders), there was no obvious difference between the leaves under different leaves of stress (primary, intermediate and advanced stress) in terms of spectral reflectivity, but the difference between these diseased leaves and normal leaves was obvious at 420-700 and 750-1 050 nm (Fig. 1).
In the four states (level 0, primary, intermediate and advanced level), there was a consistent changing trend of hyperspectral reflectivity of apple leaves. That is, there were reflection peaks at 380 and 550 nm, an absorption valley at 680 nm, red edges at 680-780 nm, and a near??infrared reflection platform after 780 nm. In the near??infrared band (780-1 050 nm), the reflectivity of normal leaves was the highest, while that of the primary damaged leaves was the lowest, and that of the advanced and intermediate damaged leaves was moderate. This indicates that the internal cell structure of normal leaves was normal, and reflection was formed many times, so the reflectivity of normal leaves was the highest. The internal cells of leaves under the stress of red spiders were destroyed, so the reflectivity of the diseased leaves was lower than that of normal leaves. As the degree of stress deepened, the reflectivity of the red spiders attached to the leaves gradually increased with the increase of band, so the reflectivity of the advanced leaves was gradually higher than that of the intermediate leaves in the near??infrared band (850-1 050 nm).
There were big differences between normal leaves and the leaves under the stress of red spiders at 420-700 and 750-1 050 nm, which can be used as the recognition bands for judging whether the leaves suffer from the stress of red spiders.
The spectral reflectivity of red spiders also showed certain vegetation characteristics. In the near??infrared band (780-1 050 nm), the reflection of red spiders led to the increase of reflectivity of leaves, which can also provide reference for the classification of red spider disease index.
In the range of 400-780 nm, there was no obvious difference between the apple leaves under the stress of red spiders in terms of spectral reflectivity.
Extraction of red edge position
The red edge is the point where the reflectivity of green plants increases fastest at 680-750 nm and the inflection point of the first derivative in this interval, which is caused by the strong absorption of plants in the red light band and the strong reflection in the near??infrared band.
Seen from Fig. 2, the changing trends of the first derivative of normal leaves and the advanced leaves were similar, and there was an obvious trough and peak at 400 and 720 nm respectively. Red edge position was extracted by using the band where the maximum of the first derivative was located. Red edge position was at 720 nm.
Correlation analysis of the spectral reflectivity and its first derivative of apple leaves under the stress of red spiders with chlorophyll content
In the early period of apply growth, there was no big difference between the primary, intermediate and advanced leaves damaged by red spiders in terms of reflectivity, so the three kinds of leaves can be called the leaves under the stress of red spiders. Here the advanced leaves damaged by red spiders were analyzed.
The correlation between the original spectral reflectivity and chlorophyll content of normal leaves was analyzed, and SPSS statistical analysis was conducted. As shown in Fig. 3, there was a negative correlation between the original spectral reflectivity and chlorophyll content of normal leaves. The negative correlation was significant (P The correlation between the original spectral reflectivity and chlorophyll content of the advanced leaves was also analyzed, and SPSS statistical analysis was conducted. Seen from Fig. 3, the correlation between the original spectral reflectivity and chlorophyll content of the advanced leaves was not significant, and the correlation coefficient was small, but there was also an obvious trough and peak.
The correlation between the first derivative of original spectral reflectivity and chlorophyll content of normal leaves was analyzed, and SPSS statistical analysis was conducted. According to Fig. 4, there was an extremely significant negative correlation between them at 513-539, 564-585, 694 (Pearson correlation coefficient was -0.822**) and 699 nm (Pearson correlation coefficient was -0.877**).
There was a significant correlation between them at 372 (Pearson correlation coefficient was -0.672*), 498 (Pearson correlation coefficient was -0.694*), 508 (Pearson correlation coefficient was -0.739*), 544 (Pearson correlation coefficient was -0.739*), 549 (Pearson correlation coefficient was -0.749*), 559 (Pearson correlation coefficient was -0.758*), 590 (Pearson correlation coefficient was 0.736*), 611 (Pearson correlation coefficient was 0.759*), 642 (Pearson correlation coefficient was 0.716*), 705 (Pearson correlation coefficient was -0.759*), 816 (Pearson correlation coefficient was 0.749*), 843 (Pearson correlation coefficient was -0.726*), 945 (Pearson correlation coefficient was 0.674*) and 951 nm (Pearson correlation coefficient was 0.771*).
The correlation between the first derivative of original spectral reflectivity and chlorophyll content of the advanced leaves was also analyzed, and SPSS statistical analysis was conducted. Seen from Fig. 4, the correlation was significantly positive at 961 nm, and Pearson correlation coefficient was 0.632*. It was significantly negative at 972 nm, and Pearson correlation coefficient was -0.723*.
Seen from the above analysis, for the normal leaves of apple, the original spectral reflectivity at 377 nm, the first derivative of original spectral reflectivity at 513-539, 564-585, 694 and 699 nm, and the original spectral reflectivity at 720 nm (red edge position) were chosen as sensitive variables.
For the advanced leaves of apple, the first derivative of original spectral reflectivity at 961 and 972 nm and the original spectral reflectivity at 720 nm (red edge position) were chosen as sensitive variables. Establishment of models based on sensitive bands
Using SPSS software, the variables in the sensitive bands were transformed to establish various regression models, and the models with higher fitness in each transformation were found (Table 1). According to Table 1, the R2 value of the linear function estimation model that was determined by using the original spectral reflectivity at 377 nm as the variable was the largest, up to 0.427*. The R2 value of the logarithmic function estimation model that was determined by using the original spectral reflectivity at 377 nm as the variable was the second largest, up to 0.425*. The two estimation models were used for further model verification.
Seen from Table 2, the R2 value of the linear function estimation model that was determined by using the first derivative of original spectral reflectivity at 972 nm as the variable was the largest, up to 0.523*. The R2 value of the linear function estimation model that was determined by using the first derivative of original spectral reflectivity at 961 nm as the variable was the second largest, up to 0.400*. The two estimation models were used for further model verification.
Model verification
To test the accuracy and reliability of the estimation models, the test data of apple leaves measured in the same group of experiments were randomly selected to test and verify the estimation models of the selected SPAD values of apple leaves, and the models with high test accuracy were screened out (Table 3 and Table 4).
Conclusions and Discussion
Influenced by natural factors such as season, soil and climate and human factors such as fertilization, cultivation techniques and management, the hyperspectral information of apple leaves will change differently. In this study, the chlorophyll content of apple leaves in Shandong was monitored, and the models were verified based on the data of samples in the same region, which enhanced the credibility and adaptability of the monitoring models. However, whether these models apply to the detection of chlorophyll content in apple leaves in different regions and growth periods needs further exploration.
In this study, the relationship between chlorophyll content and hyperspectral characteristic parameters was used to establish the estimation models of chlorophyll content in apple leaves. After the accuracy test, the best estimation model for the chlorophyll content of normal apple leaves in Shandong was determined as follows: SPAD=152.450-1 884.851 R377. The best estimation model for the chlorophyll content of the advanced apple leaves was determined as follows: SPAD=49.371-46 428.473 R??972. These models provide a method and reference for the estimation of chlorophyll content in apple leaves, which has certain guiding significance and reference value for the precision fertilization and rapid and non??destructive growth monitoring of apple. References
[1] JIANG JB, CHEN YH, HUANG WJ. Using hyperspectral remote sensing to estimate canopy chlorophyll density of wheat under yellow rust stress[J]. Spectroscopy and Spectral Analysis, 2010,30(8):2243-2247. (in Chinese).
[2] WANG JH, HUANG WJ, LAO CL, et al. Inversion of winter wheat foliage vertical distribution based on canopy reflected spectrum by partial least squares regression method[J]. Spectroscopy and Spectral Analysis, 2007,27(2):1319-1322. (in Chinese).
[3] WANG KR, PAN WC, LI SK, et al. Monitoring models of the plant nitrogen content based on cotton canopy hyperspectral reflectance[J]. Spectroscopy and Spectral Analysis, 2011,31(7):1868-1872. (in Chinese).
[4] HAN ZY, ZHU XC, FANG XY, et al. Hyperspectral estimation of apple tree canopy LAI based on SVM and RF regression[J]. Spectroscopy and Spectral Analysis, 2016,36(3):800-805. (in Chinese).
[5] HAN ZY, ZHU XC, WANG L, et al. Hyperspectral evaluation of SPAD value of apple tree canopy based on continuum??removed method[J]. Laser & Optoelectronics Progress,2016,53(2):214-223. (in Chinese).
[6] CHENG LZ, ZHU XC, GAO LL, et al. Estimation of chlorophyll content in apple leaves based on RGB model using digital camera[J]. Acta Horticulturae Sinica,2017,44(2):381-390. (in Chinese).
[7] CHENG LZ, ZHU XC, GAO LL, et al. Hyperspectral estimation of phosphorus content for apple leaves based on the random forest model[J]. Journal of Fruit Science,2016,33(10):1216-1229. (in Chinese).
[8] HU RM,WEI M, JING X, et al. Research for extracting method of apple leaf ill spots based on hyperspectral image[J]. Journal of Northwest A & F University (Natural Science Edition),2012,40(8):95-99. (in Chinese).