Method for monitoring damage degree of rice planthoppers by using unmanned aerial vehicle

Abstract

The invention provides a method for monitoring rice planthopper harm for the field of agricultural pests. In order to solve the problems that rice planthopper harm is serious and a large-area intelligent monitoring method is lacked, the invention provides a method for monitoring the rice planthopper harm degree by collecting rice multi-spectral images through an unmanned aerial vehicle, and the method is characterized in that the unmanned aerial vehicle is used for carrying a multi-spectrometer to collect the rice spectral images; according to the rate of dead leaves caused by rice planthopper damage, judging the damage degree grade of the rice in each selected plot in the image, extracting the spectral reflectivity of the rice in each plot, and constructing a spectral index; an image gray level co-occurrence matrix algorithm is adopted to extract a rice texture feature index of each plot, and a spectrum and texture index combined rice planthopper hazard degree grade monitoring model is established; performing unmanned aerial vehicle spectrum image acquisition on rice in a to-be-detected area, extracting spectrum and texture indexes, substituting the spectrum and texture indexes into the model, and monitoring the grade of the degree of damage to the rice by the rice planthoppers. The method is easy and convenient to use, high in monitoring accuracy and high in efficiency, and a new method is provided for large-area automatic monitoring of rice planthopper harm.

Description

Technical Field

[0001] The invention belongs to the field of agricultural pests, and in particular relates to a method for monitoring the degree of damage caused by important agricultural pests, rice planthoppers, by using unmanned aerial vehicles. Background Art

[0002] Rice planthoppers are destructive pests on rice, causing rice yield losses every year. Rice planthoppers absorb juice from rice plant tissues, causing the loss of nutrients and water, resulting in reduced grain plumpness. When the damage is serious, the rice plants will wither and show "louse burn" or "penetration" symptoms, resulting in no rice grains. Rapid, accurate and timely monitoring of the degree of damage to rice by rice planthoppers can not only guide field prevention and control, but also predict the level of rice yield loss in advance.

[0003] The monitoring of the degree of damage caused by rice planthoppers to rice is usually reflected by the size of the pest population. The greater the density of rice planthoppers in the field, the higher the degree of damage to the rice. It is generally believed that prevention and control is required when the number reaches 5-10 heads / cluster. Nowadays, the manual field disk photo visual method is often used to monitor the number of rice planthoppers. Professionals need to go deep into the field to take photos, identify and count at multiple points to complete the task, which is time-consuming and laborious. This method is difficult to automate the survey and is inconsistent with the needs of modern agriculture to develop in the direction of intelligence. Studies have shown that after rice planthoppers damage rice, it will cause changes in rice leaf color and plant growth, and the rate of rice leaves turning yellow will be different depending on the degree of damage caused by rice planthoppers. There are reports that the degree of damage caused by rice planthoppers is divided into 0-9 levels: 0 is not damaged; 1 is a very small number of plants with slight yellowing; 3 is partial yellowing of leaves, but no "louse burn"; 5 is severe yellowing of leaves, shrinkage and withering of plants, and 10-25% of plants are "louse burn"; 7 is more than 50% of plants withering and "louse burn", and the rest of the rice plants are severely shrunk; 9 is the death of all rice plants. Therefore, the monitoring of rice leaf characteristics can be used to determine the degree of damage caused by rice planthoppers. Although there are reports on the use of rice canopy spectra and temperature inversion to invert the size of rice planthopper populations, it is not clear whether these methods can be used to monitor the degree of damage caused by rice planthoppers in the field. The commonly used yield measurement and evaluation after the rice is yellow and ripe can obtain true and reliable results of the degree of damage caused by rice planthoppers, but the yield measurement method has obvious lag, has no practical guiding significance for the prevention and control of rice planthoppers, and the harvesting yield measurement takes a lot of manpower and is costly, so this method is difficult to use for monitoring the degree of damage caused by rice planthoppers. With the continuous maturity of drone technology and the popularization and application of its products in many fields, plant protection drones can now automatically spray rice to prevent and control pests and diseases. Therefore, it is also possible to use drone-mounted equipment to collect large-scale rice canopy characteristics, establish a relationship between the information collected by drones and the degree of damage caused by rice planthoppers, and thus realize the use of drones to automatically monitor the degree of damage caused by rice planthoppers in large areas, and guide the efficient and accurate prevention and control of rice planthoppers in rice production. Summary of the invention

[0004] The purpose of the present invention is to provide a method for monitoring the damage degree of rice planthoppers in the field by using unmanned aerial vehicles. The method uses unmanned aerial vehicles to collect rice spectral images, extract and fuse rice spectrum and texture feature indexes, establish a monitoring model for the damage degree of rice planthoppers, and achieve the purpose of using unmanned aerial vehicles to quickly, non-destructively and efficiently monitor the damage degree of rice planthoppers in the field.

[0005] The object of the present invention is achieved by the following method: selecting rice with different degrees of damage by rice planthoppers, using unmanned aerial vehicles to collect multispectral images, demarcating small areas in the image and counting the dead leaf rate caused by rice planthoppers in the small areas, and dividing the degree of damage by rice planthoppers in the small areas into different levels according to the dead leaf rate, 1-4 levels: Level 1 is 10% or less of the leaves are dry; Level 2 is 10%-25% of the leaves are dry; Level 3 is more than 50% of the leaves are dry; Level 4 is all the leaves are dry and the plant is dead. The damage level of each small area in the rice spectral image is determined according to the classification standard of each damage level, and more than 10 small areas of each damage level are demarcated. The brightness value of the multispectral image information collected by the unmanned aerial vehicle is converted into a reflectance value through radiation calibration. The images of the designated plots of each level of damage are intercepted, and the reflectance of the rice in the plots is extracted to construct the spectral index SI, such as the normalized spectral index NDVI; the gray-level co-occurrence matrix algorithm is used to extract the rice texture feature index TI of each plot, such as the texture feature dissimilarity index Diss. Combined with the SI and TI indexes, a monitoring model for the level of damage of rice planthoppers is established. Based on the monitoring model, the multispectral images of the monitored field rice are collected and the index is extracted, and the index is substituted into the model to monitor the level of damage of rice planthoppers, so as to realize the rapid and non-destructive monitoring of the damage of rice planthoppers in the field.

[0006] Effects of the invention: Compared with the currently used rice planthopper population number visual estimation method and yield loss estimation method, the invention produces the following beneficial effects: (1) The use of unmanned aerial vehicles to collect rice multispectral images is simple, fast, low-cost and highly efficient, and can complete the collection of rice in a large area in a short time; (2) It has the prospect of automation. The data collected by the unmanned aerial vehicle can be automatically transmitted to the control platform in real time remotely. The control platform automatically extracts the index and monitors the degree of damage of the rice planthopper, thereby realizing real-time automatic monitoring of the damage of the rice planthopper; (3) The unmanned aerial vehicle can automatically complete the flight and information collection by the program, and has low requirements on the professional level of the operating personnel. It has a wide range of practical applications and strong intelligence. DETAILED DESCRIPTION

[0007] The method of the present invention can be implemented in the following manner, but is not limited to this manner.

[0008] Radiometric calibration of UAV spectra: Use a radiometric calibration plate to calibrate the UAV spectrum image in the measurement area, and establish a relationship model between the brightness value and reflectivity of the spectrum image.

[0009] Acquisition of multispectral images of rice by drone: Select clear, windless and cloudless weather, and use a drone equipped with a multispectral instrument to automatically collect and save multispectral images of rice areas with different degrees of damage by rice planthoppers. The spectral images collected by the drone are transferred to the computer for the classification of plots and their degree of damage, and the extraction of spectral and texture indexes.

[0010] Determination of the degree of damage to rice in a plot: delineate an area on the spectral image as a plot, and save the geographic information raster data of the location of each plot so that the plot can be used multiple times to extract parameters and count the number of dry rice leaves in the plot. Based on the ratio of dry rice leaves in the plot, calibrate the degree of damage to rice planthoppers in each plot (level 1-4), and there should be no less than 10 plots for each level of damage.

[0011] Extraction of rice spectral index in plots: For the spectral images of each plot, the relationship model between the spectral image brightness value and the reflectivity is used to extract the spectral reflectivity at each band, such as the spectral reflectivity R at 650nm for red light and 860nm for near infrared light. 650 and R 860 , the spectral index SI of each cell is obtained, such as the normalized spectral index NDVI at 650nm red light and 860nm near infrared light. The calculation formula is NDVI 860 / 650 =(R 860 -R 650 ) / (R 860 +R 650 ).

[0012] Extraction of rice texture index in plots: The grayscale co-occurrence matrix algorithm is used to extract the rice texture index TI of each plot. The grayscale value of the spectral image of each plot of rice is reduced to 2. 8 Gray levels, set a 3×3 window, with a step size of 1 and an angle of 0, take pixel value pairs, generate a 256×256 matrix, count the number of occurrences of each pixel value pair in the co-occurrence matrix, and calculate the rice texture feature index TI of each plot in the multispectral image, such as the dissimilarity index Diss , the calculation formula is ,in, i and j Represents the grayscale level, that is, the row and column numbers of the co-occurrence matrix; P i ,j is the element in the normalized gray-level co-occurrence matrix, which represents the gray level i and j The probability of occurrence of an associated pair of pixel values; N is the number of gray levels in the image, such as 256.

[0013] Establishment of a monitoring model for the degree of rice damage: The spectral index SI and texture index TI of rice in each plot are standardized and used as variables or factors, and the degree of damage of rice in the plot (such as 1, 2, 3, 4) is used as the dependent variable for modeling, such as linear discriminant classification model, to establish a monitoring model for the degree of damage of rice planthoppers.

[0014] Model accuracy evaluation: The monitoring model is evaluated using the overall accuracy and Kappa coefficient. The calculation formula is: Overall accuracy = (true positive examples + true negative examples) / (true positive examples + false negative examples + false positive examples + true negative examples), where true positive examples refer to the number of samples correctly monitored as positive; true negative examples refer to the number of samples correctly monitored as negative; false positive examples refer to the number of samples incorrectly monitored as positive; false negative examples refer to the number of samples incorrectly monitored as negative. If the sample is actually level 1 and the monitoring is also level 1, it is a true positive example. If it is monitored as non-level 1 (level 2, 3, 4), it is a false positive example. If the sample is not actually level 1 (level 2, 3, 4), it is monitored as level 1, which is a false negative example, and if it is monitored as non-level 1, it is a true negative example. The overall accuracy ranges from 0 to 1. The closer it is to 1, the higher the accuracy. The calculation formula for Kappa is: Kappa=(overall accuracy - P e ) / (1-P e ), where P e =[(true positive examples + false negative examples)(true positive examples + false positive examples)+(false positive examples + false negative examples)(false negative examples + true negative examples)] / (true positive examples + false positive examples + true negative examples + false negative examples), P e It is the probability of agreement by chance, the probability of reaching agreement even if the evaluation is done randomly. The value range of the Kappa coefficient is between -1 and 1. Models with a Kappa value of 0.8 or above can be used for monitoring.

[0015] Monitoring of the damage degree of rice in the tested field: Use drones to collect multispectral images of the tested field rice, extract the spectral index SI and texture index TI of rice in each measured area, substitute them into the rice damage degree monitoring model, and obtain the damage degree of rice in the measured area.

[0016] Implementation effect: In this implementation example, a take-off and landing quad-rotor drone equipped with a multi-spectral camera collected drone spectral images of mature rice with a resolution of 0.95 cm at a flight altitude of 20m. The weather was clear during the measurement, and the measurement time was 14:00-15:00. Before collecting the spectral image of rice, the drone was flown above diffuse reflection white boards with reflectivity of 25%, 50% and 75%, and the entire image of each reflectivity white board was collected respectively to establish a relationship model between the brightness value and reflectivity of the spectral image, that is, radiation calibration. The drone spectral images of rice with different degrees of damage by rice planthoppers were automatically collected, and then small areas were delineated in the image to identify the level of damage by rice planthoppers in each area. Subsequently, the reflectivity of the spectral image of each area was extracted at red light 650nm and near-infrared light 860nm to obtain R 650 and R 860 , and then the normalized spectral index NDVI 860 / 650 The texture dissimilarity index Diss of the image at 860nm in the near infrared band was extracted by using the image gray-level co-occurrence matrix algorithm, and the NDVI 860 / 650 After the values ​​of NDVI and Diss are standardized to eliminate the difference in dimensions, they are used to establish a monitoring model for the degree of hazard. There are 7 samples in each of the plots with hazard levels 1, 2, 3, and 4. 860 / 650 and Diss as model factors, and the linear discriminant classification modeling method was used to establish a discriminant monitoring model for each level (Y) of rice planthopper damage in rice plots (Table 1). The model was used to monitor and verify the level of rice planthopper damage in 12 other drone spectral monitoring areas. The accuracy of the monitoring results was between 92% and 100%, and the monitoring accuracy of level 3 and level 4 plots with high rice planthopper damage reached 100%. The Kappa value of the monitoring model was between 0.91 and 1.0 (Table 1), indicating that the monitoring accuracy of this method is high, and the monitoring results are almost consistent with the actual situation, which can be used for field monitoring of rice planthopper damage.

[0017] The implementation of this method shows that the linear discriminant classification model is constructed by using multispectral images collected by drones and the extracted rice canopy spectral index and texture index as the input features of the model. The model has excellent monitoring effect on the degree of damage of rice planthoppers and can be used to monitor the damage level of rice planthoppers. This method only needs to use multispectral drones to collect images in the field. The operation is simple, the professional requirements are not high, the work intensity is not high, and it can be completed by one person; at the same time, drone collection and data transmission calculation can be automatically completed through the software platform. Therefore, the method of monitoring rice planthopper damage by drones has the prospect of automation and intelligence.

[0018] Table 1. NDVI based on drone spectral index 860 / 650 Classification model and effect of rice planthopper damage level discrimination based on near-infrared texture dissimilarity index Diss Hazard Level Discriminative classification monitoring model Accuracy% Kappa value Level 1 <![CDATA[Y=2.38×Diss+40.97×NDVI 860 / 650 -22.7]]> 92 0.91 Level 2 <![CDATA[Y=3.28×Diss+37.31×NDVI 860 / 650 -19.88]]> 92 0.91 Level 3 <![CDATA[Y=-0.91×Diss-10.27×NDVI 860 / 650 -2.78]]> 100 1 Level 4 <![CDATA[Y=-3.75×Diss-54.85×NDVI 860 / 650 -39.73]]> 100 1

Claims

1. A method for monitoring the degree of damage of rice planthoppers, characterized in that: A spectrometer mounted on an unmanned aerial vehicle was used to collect multispectral images of rice in plots with different degrees of rice planthopper damage. The proportion of dry rice leaves in the selected area of ​​the image was counted to determine the degree of rice damage caused by rice planthoppers. The reflectivity of rice in the selected area was extracted, and the spectral index SI was constructed. The gray-level co-occurrence matrix algorithm was used to extract the rice texture feature index TI in the selected area, and a monitoring model for the degree of rice planthopper damage combined with SI and TI was established. During field monitoring, unmanned aerial vehicles were used to collect multispectral images and extract the SI and TI of rice, which were substituted into the established monitoring model to obtain the degree of rice damage caused by rice planthoppers in the field.