Daytime heavy fog remote sensing identification method based on sunflower 9-waveband combination characteristics

Abstract

The invention discloses a daytime heavy fog remote sensing identification method based on sunflower 9 wave band combination characteristics, comprising the following steps: S1, downloading sunflower 9 data, and then preprocessing the data; s2, data preprocessing comprises image cutting and water area and land masking; s3, constructing and extracting cloud and haze characteristic factors; s4, training the model, comparing precision, and selecting an optimal model; in the aspect of feature extraction, it is noticed that illumination intensity, slope features between wavebands and fitting line features between the wavebands all exist, the difference is that when illumination intensity is high, the feature difference is large, and when illumination intensity is low, the feature difference is small, which becomes a practical basis for good applicability of the method. In the continuous change process of the solar altitude, the method can adapt to the change of the spectrum curve magnitude caused by the change of the illumination intensity, and has certain superiority compared with a single-band threshold value or temperature threshold value judgment method.

Description

Technical Field

[0001] The invention belongs to the technical field of daytime fog recognition, and in particular relates to a daytime fog remote sensing recognition method based on sunflower 9-band combination characteristics. Background Art

[0002] In the analysis of fog by satellite remote sensing, the focus is on the identification and extraction of fog using satellite remote sensing data. It is found that the research and analysis of the radiation characteristics of fog show that the albedo of the near-infrared band in the fog area is lower than that of the visible light band; the albedo of the visible light and near-infrared bands in the fog area is higher than that of the underlying surface, and lower than that of medium and high clouds; the infrared band brightness temperature in the fog area is usually lower than that of the underlying surface, higher than that of medium and high clouds, but in the case of weather accompanied by an inversion layer (radiation fog), the brightness temperature is higher than that of the adjacent areas of the underlying surface; the brightness temperature of the infrared band of 10.5-11.5μm in the fog area at night is higher than that of the band of 11.5-12.5μm and other spectral characteristics. Therefore, the dual spectrum method can be used to identify low-level clouds and fog at night. This method has certain effects under certain circumstances, but it has limitations in recognition accuracy. According to the research on the daytime fog identification method based on the high temporal and spatial resolution of the FY2E geostationary satellite, it was found that the reflectivity of daytime fog is generally between 20% and 50%, and the infrared brightness temperature is generally concentrated in the range of 270 to 285K. Therefore, the purpose of roughly judging the fog can be achieved by combining the dual-channel thresholds; then, by extracting the dual-channel thresholds in real time and dynamically, the threshold interval can be narrowed, which is conducive to reducing the false alarm rate; the active remote sensing data of the geostationary satellite MTSAT-1R is used to continuously monitor the foggy weather, and it is found that the formation, development and dissipation of fog can be continuously monitored day and night, which is an effective method for real-time monitoring of fog, but it is not conducive to the continuous detection of fog with high temporal and spatial resolution.

[0003] Existing technical solutions mainly focus on small scales, setting threshold ranges and other methods for fog identification. According to the research on daytime fog identification methods based on high temporal and spatial resolution conducted by the FY2E geostationary satellite, it is found that the reflectivity of daytime fog is generally between 20% and 50%, and the infrared brightness temperature is generally concentrated in the range of 270 to 285K. Therefore, the purpose of rough fog judgment can be achieved by combining dual-channel thresholds; then, by extracting dual-channel thresholds in real time and dynamically, the threshold interval can be narrowed, which is conducive to reducing the false alarm rate. This threshold method has a certain effect on a small range of temporal and spatial scales, but when applied to a large-scale spatial range, it is difficult to ensure the portability and applicability of the threshold.

[0004] To this end, we introduced a daytime fog remote sensing identification method based on the sunflower 9-band combination characteristics. Summary of the invention

[0005] The purpose of the present invention is to provide a daytime fog remote sensing identification method based on the sunflower 9-band combination characteristics. By analyzing the spectral characteristics of different types of objects on the sunflower image data, a neural network model different from the traditional single-band information as input is established. By inputting the characteristic factors constructed by the patent of the present invention, the problem of daytime fog identification in a large space is effectively solved. Through verification, the model has wider applicability and portability than conventional models, so as to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a daytime fog remote sensing identification method based on sunflower 9-band combination characteristics, comprising the following steps:

[0007] S1, first download and obtain the sunflower 9 data, and then perform data preprocessing;

[0008] S2, data preprocessing including image cropping, water and land masks;

[0009] S3, construction and extraction of cloud, fog and haze characteristic factors;

[0010] S4. Train the model, compare the accuracy, select the best model, and finally select one-third of the samples for model verification through the trained model. The model with the highest accuracy is selected as the best classification model.

[0011] Preferably, S2 includes:

[0012] The water area mask is mainly removed based on the ratio of red light to near infrared bands, and the difference greater than 1 is selected as the water area;

[0013] The land mask first calculates the cloud index, which is defined as the sum of red light and blue light divided by the near-infrared band. Here, the threshold of 1.88 is taken as the distinction between cloud haze and land.

[0014] Preferably, S3 includes:

[0015] Characteristic factor 1 is mainly the slope constructed by the b3 (0.64μm) and b5 (1.6μm) bands, as shown in the following formula:

[0016]

[0017] Preferably, rrs(b3) is the reflectivity of b3 band, rrs(b5) is the reflectivity of b5 band, b3=0.64, b5=1.6, and T1>0 mostly corresponds to foggy pixels.

[0018] Preferably, S3 includes: On the other hand, the characteristic factor 2 is constructed using the b4 (0.86 μm), b5, and b6 (2.3 μm) bands, as shown in the following formula:

[0019]

[0020] Preferably, rrs′(b5) is the reflectivity of the fitting line of b4 and b6 in the b5 band, b4=0.86, b6=2.3, and T2>0 mostly corresponds to foggy pixels.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] In terms of feature extraction, it is noted that regardless of the light intensity, the slope characteristics between the bands and the fitting line characteristics between the bands will exist. The difference is that when the light is strong, the feature difference is large, and when the light is weak, the feature difference is small. This becomes the practical basis for its good applicability. In the process of constantly changing solar altitude angle, it can adapt to the changes in the magnitude of the spectral curve caused by changes in light intensity. Compared with the single-band threshold or temperature threshold judgment method, it shows certain superiority. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flowchart of a method for remote sensing identification of daytime fog based on the sunflower 9-band combination feature provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] See also Figure 1 The present invention provides a technical solution: a daytime fog remote sensing identification method based on the sunflower 9-band combination feature, comprising the following steps:

[0026] S1, first download and obtain the sunflower 9 data, and then perform data preprocessing;

[0027] S2, data preprocessing including image cropping, water and land masks;

[0028] S3, construction and extraction of cloud, fog and haze characteristic factors;

[0029] S4. Train the model, compare the accuracy, select the best model, and finally select one-third of the samples for model verification through the trained model. The model with the highest accuracy is selected as the best classification model.

[0030] The S2 includes:

[0031] The water area mask is mainly removed based on the ratio of red light to near infrared bands, and the difference greater than 1 is selected as the water area;

[0032] The land mask first calculates the cloud index, which is defined as the sum of red light and blue light divided by the near-infrared band. Here, the threshold of 1.88 is taken as the distinguishing value between clouds, haze and land.

[0033] The S3 includes:

[0034] Characteristic factor 1 is mainly the slope constructed by the b3 (0.64μm) and b5 (1.6μm) bands, as shown in the following formula:

[0035]

[0036] Among them, rrs(b3) is the reflectivity of b3 band, rrs(b5) is the reflectivity of b5 band, b3=0.64, b5=1.6, and T1>0 mostly corresponds to foggy pixels.

[0037] The S3 includes: On the other hand, the characteristic factor 2 is constructed by using the b4 (0.86 μm), b5, and b6 (2.3 μm) bands, as shown in the following formula:

[0038]

[0039] Among them, rrs′(b5) is the reflectivity of the fitting line of b4 and b6 in the b5 band, b4=0.86μm, b6=2.3μm, and T2>0 mostly corresponds to foggy pixels.

[0040] Data acquisition and preliminary processing

[0041] Data download: First, obtain the required Sunflower 9 data through a professional data receiving system or satellite data platform. This data contains rich information about the earth's surface and is the basis for subsequent analysis. During the download process, it is necessary to ensure the integrity and accuracy of the data, covering all necessary band data in the study area and the corresponding time period;

[0042] Data preprocessing - image cropping: Since the original Sunflower 9 data covers a wide range, image cropping is required to improve processing efficiency and focus on the study area. According to the geographical location and boundary coordinates of the study area, professional remote sensing image processing software is used to accurately extract the image data of the target area from the original image and remove irrelevant surrounding areas, thereby reducing the amount of data and speeding up subsequent processing;

[0043] Data preprocessing - water and land masks

[0044] Water area mask: Water areas have unique spectral characteristics in satellite images. The ratio of red light to near-infrared bands is used to distinguish water areas from other landforms. After a large number of experiments and data analysis, it is found that when the difference between the red light and near-infrared bands is greater than 1, the area is likely to be water areas. By writing corresponding algorithms, each pixel in the image is calculated and judged, and the pixels that meet the conditions are marked as water areas, thereby generating a water area mask. In this way, in subsequent analysis, the interference of water areas can be effectively eliminated, and the focus can be on the identification of fog on land.

[0045] Land mask: In order to further accurately distinguish land from clouds and haze, it is necessary to calculate the cloud index. The calculation formula of the cloud index is the sum of red light and blue light divided by the near-infrared band. After many experiments and verifications, the threshold value of 1.88 is determined as the boundary value for distinguishing clouds, haze and land. When the calculated cloud index is greater than 1.88, the pixel is judged to be likely to belong to the cloud and haze area; otherwise, it belongs to land. In this way, a land mask is generated, laying the foundation for the subsequent accurate identification of heavy fog.

[0046] Key feature factor construction and extraction

[0047] Characteristic factor 1 is constructed based on the slope of the b3 (0.64 μm) and b5 (1.6 μm) bands. The reflectivity of these two bands is different in different types of objects, and the slope formed by this difference can reflect certain characteristics of the objects.

[0048]

[0049] Among them, rrs(b3) represents the reflectivity of the b3 band, rrs(b5) represents the reflectivity of the b5 band, b3 is fixed at 0.64μm, and b5 is fixed at 1.6μm. In foggy areas, this slope has a unique numerical range and change trend, which is significantly different from other landforms. Therefore, it can be used as one of the important bases for identifying foggy areas. In this formula, in addition to rrs(b3) and rrs(b5) representing the reflectivity of the b3 and b5 bands respectively, T1 is an important judgment parameter. Studies have found that when T1>0, the pixel mostly corresponds to foggy pixels. This characteristic factor further enhances the recognition ability of foggy areas from another perspective. Combined with characteristic factor 1, it can more comprehensively reflect the spectral characteristics of foggy areas.

[0050] Characteristic factor 2 construction (based on b4, b5, b6 bands): In order to further improve the accuracy of fog recognition, characteristic factor 2 is also constructed using b4 (0.86μm), b5 (1.6μm), and b6 (2.3μm) bands. The calculation formula is:

[0051]

[0052] Where rrs′(b5) represents the reflectivity of the fitting line between b4 and b6 at the b5 band, b4 is 0.86μm, and b6 is 2.3μm. When T2>0, the pixel mostly corresponds to the fog pixel. This characteristic factor based on multi-band construction comprehensively considers the relationship between different bands, can capture more information related to heavy fog, and improves the accuracy of recognition.

[0053] Model training and optimization selection

[0054] Model training: Collect a large amount of sample data after data preprocessing and feature factor extraction, and input this data into the pre-selected model for training. You can choose a variety of different types of models, such as neural network models, support vector machine models, etc., to compare the performance of different models when processing this data. During the training process, the model will learn the characteristics and rules in the sample data and continuously adjust its own parameters to improve its ability to identify heavy fog.

[0055] Model validation: In order to evaluate the performance and accuracy of the model, one-third of the samples were selected as the validation set. The trained model was applied to the validation set data, and the model's precision and recall rate for fog recognition were calculated. The precision rate reflects the proportion of fog samples that are truly foggy among those recognized by the model. The higher the precision rate, the better the recognition accuracy of the model.

[0056] Optimal model selection: Compare the accuracy of different models on the validation set and select the model with the highest accuracy as the final optimal classification model. This optimal model will have the best fog recognition ability and can more accurately identify foggy areas during the day in practical applications, providing reliable technical support for meteorological research, traffic warnings, etc.

[0057] References:

[0058] Reference 1:

[0059] Night fog recognition based on the new generation geostationary meteorological satellite Sunflower-8[J], author, Wang Hongbin, Zhang Zhiwei, Liu Duanyang, et al., Plateau Meteorology, 2018, 37(06): 1749-1764.

[0060] Working principle:

[0061] The slope characteristic index is constructed using the third and fifth bands measured by the visible and infrared scanning radiometers on board Sunflower-9;

[0062] The reflectance difference characteristic index of the fifth band was constructed using the fourth, fifth and sixth bands measured by the visible light and infrared scanning radiometers carried by Sunflower-9.

[0063] The KNN classification model trained with techniques 1 and 2 as input;

[0064] The same technology can be applied to the Sunflower-8 visible and infrared scanning radiometer sensors.

[0065] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A remote sensing identification method for daytime fog based on the combined features of the sunflower 9 bands, characterized in that: The following steps are involved: S1, first download and obtain the sunflower 9 data, and then perform data preprocessing; S2, data preprocessing including image cropping, water and land masks; S3, construction and extraction of cloud, fog and haze characteristic factors; S4. Train the model, compare the accuracy, select the best model, and finally select one-third of the samples for model verification through the trained model. The model with the highest accuracy is selected as the best classification model.

2. The daytime fog remote sensing identification method based on the sunflower 9-band combination feature as claimed in claim 1 is characterized in that: The S2 includes: The water area mask is mainly removed based on the ratio of red light to near infrared bands, and the difference greater than 1 is selected as the water area; The land mask first calculates the cloud index, which is defined as the sum of red light and blue light divided by the near-infrared band, and takes a threshold of 1.88 as the distinction between cloud, haze and land.

3. The daytime fog remote sensing identification method based on the sunflower 9-band combination feature as claimed in claim 1 is characterized in that: The S3 includes: Characteristic factor 1 is mainly the slope constructed by b3 and b5, as shown in the following formula: Among them, rrs(b3) is the reflectivity of b3 band, rrs(b5) is the reflectivity of b5 band, b3=0.64μm, b5=1.6μm, and T1>0 mostly corresponds to foggy pixels.

4. The daytime fog remote sensing identification method based on the sunflower 9-band combination feature as claimed in claim 1 is characterized in that: The S3 includes: using b4, b5, and b6 to construct characteristic factor 2, as shown in the following formula:

5. The daytime fog remote sensing identification method based on the sunflower 9-band combination feature as claimed in claim 4 is characterized in that: in, rrs′(b5) is the reflectivity of the fitting line between b4 and b6 in the b5 band, b4=0.86μm, b6=2.3μm, and T2>0 mostly corresponds to foggy pixels.

Images