Remote sensing identification method and equipment for plastic film mulching cultivated land based on time sequence spectral characteristics

Abstract

The invention discloses a plastic film mulching cultivated land remote sensing recognition method based on time sequence spectral characteristics, and belongs to the technical field of remote sensing ground feature information extraction. The plastic film mulching cultivated land identification index provided by the invention solves the problems that an existing plastic film mulching cultivated land identification method depends on a large number of manual labeling samples, ground objects with similar spectrums are easy to confuse, the cross-regional generalization ability is insufficient and the like; according to the plastic film mulching cultivated land identification index, by digging the specific phenological change characteristics of the plastic film mulching cultivated land in the crop growth season, the capability of distinguishing the film mulching cultivated land from similar ground objects such as non-film mulching cultivated land and plastic greenhouses is remarkably enhanced, and classification confusion is effectively reduced; according to the method, index construction and feature extraction can be realized only by depending on a small number of cultivated land samples, the real label of the plastic film mulching cultivated land does not need to be obtained, and the dependence on manual data labeling is reduced.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing ground feature information extraction technology, specifically relating to a remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics, and also relating to computer equipment. Background Technology

[0002] With the continuous improvement of agricultural intensification, plastic-mulched farmland (PMF), as an important agricultural measure to effectively increase crop yield and enhance water and fertilizer use efficiency, has been widely promoted and applied in many countries and regions around the world. However, the rapid expansion of PMF-mulched farmland has also brought increasingly serious environmental problems, especially the accumulation of PMF residues in the soil, which poses a significant threat to the soil ecosystem. Therefore, accurately grasping the spatial distribution pattern and temporal dynamics of PMF-mulched farmland is of great significance for regional agricultural planning and environmental pollution control.

[0003] Remote sensing technology has become an important means of monitoring farmland covered by plastic film, and can be mainly divided into two categories: methods based on high-resolution optical imagery and methods based on temporal medium-resolution optical imagery. Methods based on high-resolution imagery often employ deep learning semantic segmentation models (such as U-Net, SegNet, GCN, etc.), which can identify land cover features such as plastic film and plastic greenhouses by extracting spatial texture features and contextual information. For example, some studies have achieved effective differentiation between plastic greenhouses and farmland covered by plastic film using deep learning models that include convolutional layers and non-local attention mechanisms. These methods can achieve high-precision identification of farmland covered by plastic film in small-scale areas, but they heavily rely on sub-meter or even higher resolution remote sensing data, which is costly to acquire and has limited coverage. Therefore, this method is difficult to meet the practical application requirements of identifying farmland covered by plastic film on a large scale and across multiple time periods. Furthermore, high-resolution imagery has a long revisit period, making it difficult to fully extract the temporal series features of crops and effectively distinguish other covering land cover features with similar spectral characteristics to farmland covered by plastic film, such as plastic greenhouses.

[0004] In contrast, temporal medium-resolution imagery (imagery with spatial resolution ranging from 10 m to 30 m) offers advantages such as wide coverage, short revisit periods, and rich phenological information, making it a crucial data source for identifying large-scale plastic film-covered farmland. Existing research has used medium-resolution imagery to identify plastic film-covered farmland through traditional supervised classification models (such as support vector machines, decision trees, and random forests) or deep learning models (such as long short-term memory networks). However, these methods typically require a large number of field-collected or manually labeled training samples and significant computational resources, and suffer from insufficient spatiotemporal transferability, making them unsuitable for adapting to the dynamic identification needs of plastic film-covered farmland in different regions and years. Besides supervised classification model-based methods, index-based classification using spectral indices has gained widespread attention in recent years due to its strong interpretability, low sample dependence, and good regional generalization ability. Some studies have constructed spectral indices for plastic film identification, such as decision tree rules based on near-infrared, blue light, short-wave infrared, and NDVI, or designed the Plastic-Mulched Landcover Index (PMLI) specifically for identifying plastic film-covered farmland. However, existing index methods still have two significant limitations: first, they require frequent adjustment of thresholds for different regions and years, lacking spatiotemporal generalization; second, they cannot fully capture the typical phenological processes of mulched farmland, resulting in low identification accuracy when distinguishing non-target land features with similar spectra (such as non-mulched farmland and plastic greenhouses). Summary of the Invention

[0005] The purpose of this invention is to address the aforementioned problems in the existing technology by providing a remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics, and also to provide computer equipment.

[0006] The above-mentioned objectives of the present invention are achieved by the following technical means: A remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics includes the following steps: Step 1: Obtain the remote sensing time-series image dataset of the study area, and perform preprocessing and band extraction on the remote sensing time-series image dataset in sequence, then resample to the same resolution, and finally perform median synthesis to obtain the time-series spectral dataset. Step 2: Calculate the normalized vegetation index based on the time-series spectral dataset. and Normalized Dioxide Index The time-series spectral and vegetation index datasets were calculated, and time-series smoothing filters were applied to different bands in the datasets to generate filtered time-series spectral and vegetation index datasets. The normalized filtered vegetation index is... The normalized filtered water index is ; Step 3: Obtain multiple farmland samples from the study area, and calculate the average mulch film coverage time within the study area based on these samples. Based on average mulch film covering time Normalized Filtered Vegetation Index Normalized Filtered Water Index Constructing an identification index for farmland covered by plastic film Finally, the identification index of cultivated land with plastic film mulching was performed on the time-series spectral and vegetation index filtering dataset. The calculation yielded the identification results of farmland covered by plastic film.

[0007] The remote sensing time-series image data for the study area, as mentioned above, includes Sentinel-2, Landsat-7, and Landsat-8 remote sensing time-series image data.

[0008] The time-series spectral dataset described above is obtained through the following steps: Step S1.2.1: Perform cloud removal processing on the Sentinel-2 remote sensing time-series image data to obtain cloud-removed Sentinel-2 remote sensing time-series image data; Step S1.2.2: Remove clouds, cloud shadows, snow cover, and missing stripe areas from Landsat-7 and Landsat-8 remote sensing time-series image data to obtain cloud-free Landsat-7 and cloud-free Landsat-8 remote sensing time-series image data. Step S1.2.3: The least squares regression method is used to perform band correction and alignment on the spectral reflectance of the cloud-free Sentinel-2 remote sensing time-series image data, the cloud-free Landsat-7 remote sensing time-series image data, and the cloud-free Landsat-8 remote sensing time-series image data to obtain the corrected Sentinel-2 remote sensing time-series image data, the corrected Landsat-7 remote sensing time-series image data, and the corrected Landsat-8 remote sensing time-series image data. Step S1.2.4: Extract the six bands of blue light, green light, red light, near-infrared light, short-wave infrared light 1, and short-wave infrared light 2 from the corrected Sentinel-2 remote sensing time-series image data, the corrected Landsat-7 remote sensing time-series image data, and the corrected Landsat-8 remote sensing time-series image data, and resample them to the same resolution to obtain resampled Sentinel-2 remote sensing time-series image data, resampled Landsat-7 remote sensing time-series image data, and resampled Landsat-8 remote sensing time-series image data; Step S1.2.5 Finally, the median synthesis method is applied to the resampled Sentinel-2 remote sensing time-series image data, the resampled Landsat-7 remote sensing time-series image data, and the resampled Landsat-8 remote sensing time-series image data to obtain synthetic remote sensing time-series image data. All synthetic remote sensing time-series image data constitute a time-series spectral dataset.

[0009] As described above, step 2 specifically includes the following steps: Step S2.1: Calculate the normalized vegetation index for each synthetic remote sensing time-series image in the time-series spectral dataset. and Normalized Dioxide Index The time-series spectral and vegetation index datasets corresponding to the time-series spectral datasets are obtained. Each scene of time-series spectral and vegetation index data includes data on blue light, green light, red light, near-infrared light, short-wave infrared light 1, short-wave infrared light 2, normalized vegetation index, and normalized water index. Step S2.2: Perform time-series smoothing filtering on different bands in the time-series spectral and vegetation index dataset to obtain a filtered time-series spectral and vegetation index dataset. Each scene of the filtered time-series spectral and vegetation index data includes filtered data for blue light, green light, red light, near-infrared light, shortwave infrared light 1, and shortwave infrared light 2, as well as a normalized filtered vegetation index. and normalized filtered water index .

[0010] Normalized Difference Vegetation Index and Normalized Dioxide Index Calculated based on the following formulas respectively: ; ; In the formula, For near-infrared reflectivity, Reflectivity in the red light band This refers to the reflectivity in the green light band.

[0011] As mentioned above, the identification index of cultivated land covered by plastic film Constructed based on the following formula: ; In the formula, To determine if the target pixel is showing an upward signal, To determine if the target pixel presents a smooth signal; Calculated based on the following formula: ; In the formula, This represents the average duration of plastic film mulching. The average mulch film coverage time in the time-series spectral dataset The date of the composite remote sensing time-series image data of the next scene. For target pixels in the time-series spectral and vegetation index filtered dataset, the mean time of plastic film coverage is... Normalized filtered vegetation index for the given date value, For target pixels in time-series spectral and vegetation index filtering datasets Normalized filtered vegetation index for the given date value; Calculated based on the following formula: ; ; In the formula, For the target pixel in the time-series spectral and vegetation index filtered dataset Normalized Filtered Water Index in Temporal Spectrum and Vegetation Index Filtered Data value, This represents the total number of images in the time-series spectral and vegetation index filtered dataset. The image number is the image sequence number in the time-series spectral and vegetation index filtered dataset. , These are intermediate parameters.

[0012] As described above, the average time of mulch film coverage Specifically, it is calculated in the following way: Step 3.1: Obtain the basic farmland mask image of the study area, and select from the basic farmland mask image. A sample of cultivated land, based on The coordinates of each cultivated land sample were constructed using time-series spectral and vegetation index filtered datasets, respectively. Normalized filtered vegetation index for individual cultivated land samples Time series plot and normalized filtered water index The timing diagram; Step 3.2, based on Normalized filtered vegetation index for individual cultivated land samples The time series plots were used to calculate the average duration of plastic film mulching for all farmland samples. Average mulch film covering time Calculated based on the following formula: ; ; In the formula, The number of cultivated land samples, This refers to the serial number of the cultivated land sample. , for and The date at the third quantile, For time-series spectral and vegetation index filtering datasets, Normalized filtered vegetation index of individual cultivated land samples before March The date corresponding to the minimum value. For time-series spectral and vegetation index filtering datasets, Normalized filtered vegetation index of individual cultivated land samples from March to September The date corresponding to the maximum value.

[0013] As described above, the cultivated land samples in step 3.1 satisfy the following constraints: normalized filtered vegetation index before March. The maximum value is greater than or equal to the first set value, and the normalized filtered vegetation index between March and September The minimum value is less than or equal to the second set value.

[0014] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the remote sensing identification method for mulched farmland based on time-series spectral characteristics as described above.

[0015] Compared with the prior art, the present invention has the following advantages: (1) The arable land identification index with plastic film mulching constructed in this invention It is a specialized spectral index for identifying farmland covered by plastic film mulch. By exploring the unique phenological change trends of mulched farmland, the ability to distinguish it from spectrally similar land features (such as unmulched farmland and plastic greenhouses) has been significantly improved, effectively suppressing classification confusion.

[0016] (2) The method of the present invention can realize the identification index of cultivated land covered by plastic film by relying on only a small number of cultivated land samples. The construction and feature extraction do not require real labeled samples of farmland covered by plastic film, which greatly reduces the dependence on manually labeled data. It has good practicality and scalability, and is especially suitable for the identification of farmland covered by plastic film in areas with scarce samples or in historical years.

[0017] (3) The arable land identification index of the present invention In verification experiments conducted in several typical mulched planting areas, it demonstrated excellent spatiotemporal generalization ability. Its recognition accuracy was significantly better than traditional vegetation indices, machine learning classification models, and publicly available data products in different regions and years. It also exhibited good cross-regional mobility and stability, making it suitable for remote sensing mapping of mulched farmland over a wide area and over multiple years. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of eight types of field samples in the first study area of ​​this invention; Figure 3 This is a schematic diagram of eight types of field samples from the second research area of ​​this invention; Figure 4 This is a schematic diagram of eight types of field samples in study area three of this invention; Figure 5 These are spectral reflectance diagrams of two types of cultivated land: mulched farmland and greenhouse farmland, during the sowing period (February 20 to February 30) of Embodiment 1 of the present invention. Figure 6 These are spectral reflectance diagrams of two types of farmland: mulched farmland and greenhouse farmland, during the growing season (May 21 to June 1) of Embodiment 1 of the present invention. Figure 7 These are spectral reflectance diagrams of two types of farmland: mulched farmland and greenhouse farmland, during the peak flowering period (July 20 to July 30) of Embodiment 1 of the present invention. Figure 8 This invention relates to three types of farmland: farmland with plastic film mulching, greenhouses, and farmland without plastic film mulching, as described in Embodiment 1 of the present invention. Time-series spectral characteristic curves; Figure 9 This invention relates to three types of farmland: farmland with plastic film mulching, greenhouses, and farmland without plastic film mulching, as described in Embodiment 1 of the present invention. Time-series spectral characteristic curves; Figure 10 This is a comparison chart showing the overall accuracy OA index results of study areas one, two, and three of Embodiment 1 of the present invention, respectively, obtained by using the method of the present invention and by using other methods; Figure 11 This is a comparison chart showing the results of the F1 score index obtained by using the method of the present invention and other methods in study area 1, study area 2, and study area 3 of Embodiment 1 of the present invention, respectively. Detailed Implementation

[0019] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to embodiments. The embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] Example 1: like Figure 1 As shown, the remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics includes the following steps: Step S1: Obtain a medium-resolution remote sensing time-series image dataset of the study area using the GEE platform (Google Earth Engine). The acquired remote sensing time-series image dataset is then preprocessed and bands extracted sequentially, resampled to the same resolution, and finally synthesized using median values ​​to generate a time-series spectral dataset. This specifically includes the following steps: In this embodiment, three types of medium-resolution remote sensing time-series image data (images with spatial resolution in the range of 10 m to 30 m) were acquired, namely Sentinel-2 remote sensing time-series image data, Landsat-7 remote sensing time-series image data, and Landsat-8 remote sensing time-series image data.

[0021] Step S1.1: Download Sentinel-2 remote sensing time-series image data, Landsat-7 remote sensing time-series image data, and Landsat-8 remote sensing time-series image data through the GEE platform.

[0022] Based on the GEE platform, this invention acquires surface reflectance products from the Sentinel-2 satellite multispectral instrument (MSI), Landsat-7 satellite enhanced thematic mapper (ETM+), and Landsat-8 satellite operational land imager (OLI) for the study area in 2020 and 2024, respectively, to obtain Sentinel-2 remote sensing time-series image data, Landsat-7 remote sensing time-series image data, and Landsat-8 remote sensing time-series image data. To eliminate the interference of the freeze-thaw period, this embodiment only uses images from March 20 to October 31 of each year for synthesis.

[0023] Step S1.2: The Sentinel-2, Landsat-7, and Landsat-8 remote sensing time-series image data are preprocessed and bands extracted sequentially, then resampled to the same resolution, and finally synthesized using 10-day median values ​​to generate a time-series spectral dataset. This process includes the following steps: Step S1.2.1: First, use the cloud score product and set a threshold (in this embodiment, the threshold is set to 0.65) to perform cloud removal processing on the Sentinel-2 remote sensing time-series image data to obtain cloud-removed Sentinel-2 remote sensing time-series image data.

[0024] Step S1.2.2: Next, the CFMASK algorithm (C Function of Mask, a cloud and cloud shadow detection algorithm) is used to process the Landsat-7 and Landsat-8 remote sensing time-series image data to remove clouds, cloud shadows, snow cover, and missing strip areas, resulting in cloud-free Landsat-7 and cloud-free Landsat-8 remote sensing time-series image data.

[0025] Step S1.2.3: Then, the least squares regression method is used to perform band correction and alignment on the spectral reflectance of the cloud-free Sentinel-2 remote sensing time-series image data, the cloud-free Landsat-7 remote sensing time-series image data, and the cloud-free Landsat-8 remote sensing time-series image data, to obtain the corrected Sentinel-2 remote sensing time-series image data, the corrected Landsat-7 remote sensing time-series image data, and the corrected Landsat-8 remote sensing time-series image data.

[0026] Step S1.2.4: Next, band extraction and resampling are performed: the corrected Sentinel-2 remote sensing time-series image data, the corrected Landsat-7 remote sensing time-series image data, and the corrected Landsat-8 remote sensing time-series image data are subjected to extraction of six bands: blue, green, red, near-infrared (NIR), short-wave infrared 1 (SWIR1), and short-wave infrared 2 (SWIR2), and then uniformly resampled to a resolution of 10 meters to obtain resampled Sentinel-2 remote sensing time-series image data, resampled Landsat-7 remote sensing time-series image data, and resampled Landsat-8 remote sensing time-series image data.

[0027] Step S1.2.5: Finally, perform 10-day median synthesis: Apply the 10-day median synthesis method (i.e., perform median synthesis at 10-day intervals) to the resampled Sentinel-2 remote sensing time-series image data, the resampled Landsat-7 remote sensing time-series image data, and the resampled Landsat-8 remote sensing time-series image data respectively to obtain synthesized remote sensing time-series image data. All synthesized remote sensing time-series image data constitute a stable time-series spectral dataset.

[0028] Step S2: Calculate the normalized vegetation index based on the time-series spectral dataset. and Normalized Dioxide Index The calculation yields a time-series spectral and vegetation index dataset. The HANTS algorithm is then used to perform temporal smoothing filtering on different bands within the dataset, generating a filtered time-series spectral and vegetation index dataset. The specific steps include: Step S2.1: Calculate the normalized vegetation index for each synthetic remote sensing time-series image in the time-series spectral dataset. (Normalized Difference Vegetation Index) and Normalized Water Index (Normalized Difference Water Index) yields the time-series spectral and vegetation index datasets corresponding to the time-series spectral dataset. Each scene's time-series spectral and vegetation index data includes data from eight bands: blue light, green light, red light, near-infrared light, shortwave infrared 1, shortwave infrared 2, normalized vegetation index, and normalized water index.

[0029] Among them, the normalized vegetation index and Normalized Dioxide Index Calculated based on the following formulas respectively; (1); (2); in, For near-infrared reflectivity, Reflectivity in the red light band This refers to the reflectivity in the green light band.

[0030] Step S2.2: Use the HANTS algorithm to perform temporal smoothing filtering on different bands in the time-series spectral and vegetation index dataset to generate a time-series spectral and vegetation index filtered dataset. Each scene of the time-series spectral and vegetation index filtered data includes filtered data for blue light, green light, red light, near-infrared light, shortwave infrared light 1, and shortwave infrared light 2, as well as a normalized filtered vegetation index. and normalized filtered water index .

[0031] To remove the influence of abnormal noise in remote sensing images under cloudy and rainy conditions and improve the continuity and interpretability of time-series data, this invention uses the HANTS algorithm (Harmonic Analysis of Time Series) filtering method to perform time-series smoothing filtering on the time-series spectral and vegetation index datasets generated in step S2.1, generating a time-series spectral and vegetation index filtered dataset. The HANTS algorithm suppresses outliers and preserves vegetation phenological change characteristics by fitting the periodic characteristics of the time series. The time-series spectral and vegetation index filtered dataset serves as the basic input features for the subsequent construction of the cultivated land index under plastic film mulching.

[0032] Step S3: Obtain multiple field farmland samples from the study area, and calculate the average mulch film coverage time within the study area based on these samples. Based on average mulch film covering time Normalized Filtered Vegetation Index Normalized Filtered Water Index Constructing an identification index for farmland covered by plastic film Finally, the identification index of cultivated land with plastic film mulching was performed on the time-series spectral and vegetation index filtering dataset. The calculation yields the identification result map of farmland covered by plastic film, specifically including the following steps: This invention first collected field samples of mulched farmland and other types of samples in the study area in 2020. Based on publicly available Google 2020 meter-level remote sensing images of the Earth, the mulched farmland samples and other types of samples were visually interpreted and screened according to texture features. Erroneous field samples were eliminated, and the remaining field samples were used to construct a high-quality sample set. The high-quality sample set was then randomly divided into a training sample set and a validation sample set in a 7:3 ratio.

[0033] Other types of samples in this embodiment include greenhouses, unmulched farmland, roads, buildings, bare soil, forests, and water bodies; Figures 2-4 These are schematic diagrams of eight types of field samples from Study Area 1, Study Area 2, and Study Area 3.

[0034] Phenological characteristics of mulched farmland samples were extracted: Based on the coordinates of each training sample, spectral reflectance data of different bands of mulched farmland samples and other types of samples were extracted from the time-series spectral and vegetation index filtered dataset. The time-series spectral and vegetation index characteristics at different periods included blue light, green light, red light, near-infrared light, shortwave infrared light 1, shortwave infrared light 2, and normalized filtered vegetation index. and normalized filtered water index This embodiment plots spectral reflectance data for different bands of samples from mulched farmland, greenhouses, unmulched farmland, roads, buildings, bare soil, forests, and water bodies, such as... Figures 5-9 The following are some examples, in which Figure 5 , Figure 6 and Figure 7 The spectral reflectance of two types of farmland—land covered with plastic film and greenhouse—is shown for the sowing period (composite images from February 20 to February 30), the growing period (composite images from May 21 to June 1), and the full bloom period (composite images from July 20 to July 30). Figure 8 and Figure 9 These represent three types of farmland: farmland with plastic film mulch, greenhouses, and farmland without plastic film mulch. Time-series spectral characteristic curves and Time-series spectral characteristic curves.

[0035] Through comparative analysis, this invention uses two dual-time-series spectral features—rising signals and smoothing signals—to construct an identification index for farmland covered by plastic film. ,exist The date range corresponding to the rising signal is determined from the time-series spectral characteristic curve. The date range corresponding to the smoothed signal is determined from the time-series spectral characteristic curve.

[0036] Step S3.1: Using the publicly available farmland product results from the European Space Agency World Cover project 2020 (WorldCover) in 2020, generate a basic farmland mask map within the study area, and select from the basic farmland mask image. A sample of cultivated land, based on The coordinates of each cultivated land sample were constructed using time-series spectral and vegetation index filtered datasets, respectively. Normalized filtered vegetation index for individual cultivated land samples Time series plot and normalized filtered water index The timing diagram.

[0037] Each cultivated land sample was selected using the following method: normalized filtered vegetation index prior to March. The maximum value is greater than or equal to 0.5 (first set value), and the normalized filtered vegetation index is within the growing season phenological window (March to September). The minimum value is less than or equal to 0.2 (second set value). Samples that do not meet these threshold standards will not participate in the subsequent exponential threshold extraction.

[0038] The WorldCover dataset provides a 10-meter resolution global land cover product for 2020. This product classifies land use types based on Sentinel-1 and Sentinel-2 remote sensing satellite data and is widely used in remote sensing identification research of various land cover types.

[0039] Step 3.2: In this embodiment, 300 farmland samples are randomly selected from each study area. The farmland covered by plastic film is identified by comparing the masked farmland samples with the plastic film-covered farmland samples. Threshold extraction based on normalized filtered vegetation index from 300 farmland samples. The time series plots were used to analyze the temporal vegetation index changes under different crops or farming patterns, and to extract the phenological characteristics of cultivated land covered by plastic film, namely the average plastic film covering time of all cultivated land samples in the study area. Climbing signal time) extraction, average mulch film coverage time Calculated using the following formula: (3); (4); in, The number of cultivated land samples, This refers to the serial number of the cultivated land sample. Represents the time-series spectral and vegetation index filtered dataset for cultivated land samples ( Normalized filtered vegetation index before March The date corresponding to the minimum value. For time-series spectral and vegetation index filtering datasets, Normalized filtered vegetation index of individual cultivated land samples during the growing season phenological window (March to September) The date corresponding to the maximum value. for and The date at the one-third quartile, when the climbing signal is clearest, is the average time for mulch film coverage. represent The average date on which the climbing signal is clearest for each farmland sample.

[0040] Step 3.3: Based on average mulch film covering time Normalized Filtered Vegetation Index Normalized Filtered Water Index Constructing an identification index for farmland covered by plastic film Finally, the identification index of cultivated land with plastic film mulching was performed on the time-series spectral and vegetation index filtering dataset. The calculation yielded a map showing the identification results of farmland covered by plastic film, and the identification index of farmland covered by plastic film. Calculated based on the following formula: (5); (6); (7); (8); in, The average mulch film coverage time in the time-series spectral dataset The date of the subsequent composite remote sensing time-series image data, in this embodiment, is the average time of mulch film coverage. Add 10 days, For target pixels in the time-series spectral and vegetation index filtered dataset, the mean time of plastic film coverage is... Normalized filtered vegetation index for the given date value, For target pixels in time-series spectral and vegetation index filtering datasets Normalized filtered vegetation index for the given date value; This represents the total number of images in the time-series spectral and vegetation index filtered dataset. The image number is the image sequence number in the time-series spectral and vegetation index filtered dataset. , To statistically analyze the target pixel in the first... Scene time-series spectral and vegetation index filtered data intermediate parameters of the value, For the target pixel in the time-series spectral and vegetation index filtered dataset Normalized Filtered Water Index in Temporal Spectrum and Vegetation Index Filtered Data value.

[0041] The mulch film arable land identification index of the present invention It consists of two parts: Used to determine whether the target pixel exhibits an ascending signal (where, (To determine if the target pixel presents an ascending signal) Used to determine whether the target pixel presents a smooth signal (where, To determine if a target pixel presents a smooth signal, pixels that meet the intersection condition are identified as pixels of farmland covered by plastic film, while pixels that do not meet the intersection condition are identified as pixels of farmland not covered by plastic film.

[0042] Finally, accuracy verification was performed. Specifically, to evaluate the accuracy of the method, this invention used a validation sample set and evaluated the prediction results of the test data using overall accuracy (OA) and F1 score. The accuracy was also compared with other methods such as the Physical Mulch Coverage Index (PMLI), Random Forest (RF), Support Vector Machine (SVM), Long Short-Term Memory Network (LSTM), and the publicly available Physical Mulch Coverage Farmland Dataset (CP). The final results are as follows: Figure 10 and Figure 11 As shown in the figure, the identification index for cultivated land covered by plastic film proposed in this invention can be seen. In all study areas, it achieved recognition results superior to traditional methods, with an average OA of 84.15% and an F1 score of 81.07%, significantly outperforming the mulch film coverage index PMLI (54.42%, 45.97%), random forest RF (80.87%, 76.92%), support vector machine SVM (82.58%, 77.61%), long short-term memory network LSTM (79.21%, 74.82%), and mulch film-covered farmland dataset CP (79.85%, 60.08%).

[0043] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0044] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0045] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0046] It should be noted that the embodiments described in this invention are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains can make various modifications or additions to the described embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics, characterized in that, Includes the following steps: Step 1: Obtain the remote sensing time-series image dataset of the study area, and perform preprocessing and band extraction on the remote sensing time-series image dataset in sequence, then resample to the same resolution, and finally perform median synthesis to obtain the time-series spectral dataset. Step 2: Calculate the normalized vegetation index based on the time-series spectral dataset. and Normalized Dioxide Index The time-series spectral and vegetation index datasets were calculated, and time-series smoothing filters were applied to different bands in the datasets to generate filtered time-series spectral and vegetation index datasets. The normalized filtered vegetation index is... The normalized filtered water index is ; Step 3: Obtain multiple farmland samples from the study area, and calculate the average mulch film coverage time within the study area based on these samples. Based on average mulch film covering time Normalized Filtered Vegetation Index Normalized Filtered Water Index Constructing an identification index for farmland covered by plastic film Finally, the identification index of cultivated land with plastic film mulching was performed on the time-series spectral and vegetation index filtering dataset. The calculation yielded the identification results of farmland covered by plastic film.

2. The remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics according to claim 1, characterized in that, The remote sensing time-series image data of the study area includes Sentinel-2 remote sensing time-series image data, Landsat-7 remote sensing time-series image data, and Landsat-8 remote sensing time-series image data.

3. The remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics according to claim 2, characterized in that, The time-series spectral dataset is obtained through the following steps: Step S1.2.1: Perform cloud removal processing on the Sentinel-2 remote sensing time-series image data to obtain cloud-removed Sentinel-2 remote sensing time-series image data; Step S1.2.2: Remove clouds, cloud shadows, snow cover, and missing stripe areas from Landsat-7 and Landsat-8 remote sensing time-series image data to obtain cloud-free Landsat-7 and cloud-free Landsat-8 remote sensing time-series image data. Step S1.2.3: The least squares regression method is used to perform band correction and alignment on the spectral reflectance of the cloud-free Sentinel-2 remote sensing time-series image data, the cloud-free Landsat-7 remote sensing time-series image data, and the cloud-free Landsat-8 remote sensing time-series image data to obtain the corrected Sentinel-2 remote sensing time-series image data, the corrected Landsat-7 remote sensing time-series image data, and the corrected Landsat-8 remote sensing time-series image data. Step S1.2.4: Extract the six bands of blue light, green light, red light, near-infrared light, short-wave infrared light 1, and short-wave infrared light 2 from the corrected Sentinel-2 remote sensing time-series image data, the corrected Landsat-7 remote sensing time-series image data, and the corrected Landsat-8 remote sensing time-series image data, and resample them to the same resolution to obtain resampled Sentinel-2 remote sensing time-series image data, resampled Landsat-7 remote sensing time-series image data, and resampled Landsat-8 remote sensing time-series image data; Step S1.2.5 Finally, the median synthesis method is applied to the resampled Sentinel-2 remote sensing time-series image data, the resampled Landsat-7 remote sensing time-series image data, and the resampled Landsat-8 remote sensing time-series image data to obtain synthetic remote sensing time-series image data. All synthetic remote sensing time-series image data constitute a time-series spectral dataset.

4. The remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics according to claim 3, characterized in that, Step 2 specifically includes the following steps: Step S2.1: Calculate the normalized vegetation index for each synthetic remote sensing time-series image in the time-series spectral dataset. and Normalized Dioxide Index The time-series spectral and vegetation index datasets corresponding to the time-series spectral datasets are obtained. Each scene of time-series spectral and vegetation index data includes data on blue light, green light, red light, near-infrared light, short-wave infrared light 1, short-wave infrared light 2, normalized vegetation index, and normalized water index. Step S2.2: Perform time-series smoothing filtering on different bands in the time-series spectral and vegetation index dataset to obtain a filtered time-series spectral and vegetation index dataset. Each scene of the filtered time-series spectral and vegetation index data includes filtered data for blue light, green light, red light, near-infrared light, shortwave infrared light 1, and shortwave infrared light 2, as well as a normalized filtered vegetation index. and normalized filtered water index .

5. The remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics according to claim 4, characterized in that, Normalized Difference Vegetation Index and Normalized Dioxide Index Calculated based on the following formulas respectively: ； ； In the formula, For near-infrared reflectivity, For red light band reflectivity, This refers to the reflectivity in the green light band.

6. The remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics according to claim 1, characterized in that, The identification index of farmland covered by plastic film Constructed based on the following formula: ； In the formula, To determine if the target pixel is showing an upward signal, To determine if the target pixel presents a smooth signal; Calculated based on the following formula: ； In the formula, This represents the average duration of plastic film mulching. The average mulch film coverage time in the time-series spectral dataset The date of the composite remote sensing time-series image data of the next scene, For target pixels in the time-series spectral and vegetation index filtered dataset, the mean plastic film coverage time is... Normalized filtered vegetation index for the given date value, For target pixels in time-series spectral and vegetation index filtering datasets Normalized filtered vegetation index for the given date value; Calculated based on the following formula: ； ； In the formula, For the target pixel in the time-series spectral and vegetation index filtered dataset Normalized Filtered Water Index in Temporal Spectrum and Vegetation Index Filtered Data value, This represents the total number of images in the time-series spectral and vegetation index filtered dataset. The image number is the image sequence number in the time-series spectral and vegetation index filtered dataset. , These are intermediate parameters.

7. The remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics according to claim 6, characterized in that, The average mulch film coverage time Specifically, it is calculated in the following way: Step 3.1: Obtain the basic farmland mask image of the study area, and select from the basic farmland mask image. A sample of cultivated land, based on The coordinates of each cultivated land sample were constructed using time-series spectral and vegetation index filtered datasets, respectively. Normalized filtered vegetation index for individual cultivated land samples Time series plot and normalized filtered water index Timing diagram; Step 3.2, based on Normalized filtered vegetation index for individual cultivated land samples The time series plots were used to calculate the average duration of plastic film mulching for all farmland samples. Average mulch film covering time Calculated based on the following formula: ； ； In the formula, The number of cultivated land samples, This refers to the serial number of the cultivated land sample. , for and The date at the third quantile, For time-series spectral and vegetation index filtering datasets, Normalized filtered vegetation index of individual cultivated land samples before March The date corresponding to the minimum value. For time-series spectral and vegetation index filtering datasets, Normalized filtered vegetation index of individual cultivated land samples from March to September The date corresponding to the maximum value.

8. The remote sensing identification method for cultivated land covered by plastic film based on time-series spectral characteristics according to claim 7, characterized in that, The cultivated land samples in step 3.1 meet the following constraints: normalized filtered vegetation index before March. The maximum value is greater than or equal to the first set value, and the normalized filtered vegetation index between March and September The minimum value is less than or equal to the second set value.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the remote sensing identification method for mulched farmland based on time-series spectral characteristics as described in any one of claims 1 to 8.

Images