Oil tea planting area monitoring method and system considering forest age and growth suitability

Abstract

The invention discloses a camellia oleifera planting area monitoring method and system considering forest age and growth suitability. Comprising the following steps: extracting multi-level features of newly-built and old camellia oleifera forests from a high-resolution image and a multispectral image by using a deep learning model and a machine learning model respectively, and outputting distribution results of the multi-level features; obtaining and integrating meteorological reanalysis data and topographic feature data, calculating and normalizing suitability scores of all dimensions in combination with camellia oleifera growth suitability evaluation, and generating a camellia oleifera growth suitability distribution map; new and old camellia oleifera forest distribution results are fused into a preliminary camellia oleifera planting area, constraint is applied by means of a suitability distribution diagram, and finally a corrected camellia oleifera planting area distribution diagram is obtained. The scheme is high in detection precision and suitable for large-scale oil tea resource monitoring and management.

Description

Technical Field

[0001] The present invention belongs to the fields of remote sensing geographic information systems and computers, and more specifically, relates to a monitoring method and system for oil-tea camellia planting areas considering forest age and growth suitability. Background Art

[0002] With the development of the national economic level, the consumption capacity of Chinese people has been continuously improved, and the market space for edible vegetable oils has been continuously expanded. At present, the oil-tea camellia industry has become an important pillar of the agricultural economy in many places, and the planting area of oil-tea camellia in China has also shown an increasing trend year by year. Therefore, carrying out large-scale and accurate monitoring of oil-tea camellia planting areas is a key measure to effectively improve the utilization efficiency of land resources and is of great significance to promoting the sustainable development of the oil-tea camellia industry.

[0003] Although traditional manual field surveys can obtain accurate oil-tea camellia planting attribute information, the time and economic costs are high. In contrast, remote sensing data has the characteristics of wide coverage area and short revisit cycle, and can quickly and accurately obtain large-scale agricultural and forest land information, thus providing strong support for resource surveys and production assessments. In recent years, related research has applied unmanned aerial vehicle and satellite remote sensing data to tasks such as growth prediction, pest and disease monitoring, and rapid yield estimation of oil-tea camellia forests. Some research uses single-modal remote sensing image data, such as high-resolution optical images or multi-spectral images, for oil-tea camellia forest extraction and has made certain progress. However, there are problems with large intra-class feature differences and small inter-class differences in oil-tea camellia forests. Specifically, on high-resolution optical images, the texture features of newly planted oil-tea camellia forests are obvious, but the visual features of old oil-tea camellia forests are more similar to those of other forest lands; on medium-resolution multi-spectral or hyperspectral images, the spectral feature differences between newly planted oil-tea camellia forests and old oil-tea camellia forests are also large. This makes the current methods based on single-modal data still have poor extraction accuracy for oil-tea camellia forests.

[0004] The invention patent with the publication number CN117994701A in the prior art proposes a method for rapidly predicting the yield of oil-tea camellia based on deep learning. The method includes collecting RGB images and videos of multiple fruit-bearing oil-tea camellia trees; manually annotating the oil-tea camellia fruits in the collected images and videos to construct a target detection dataset; optimizing the oil-tea camellia fruit target detection model by adjusting the YOLOv8 model structure, and training the target detection network model using the processed dataset; performing target detection on each frame of the image by the trained network model; associating each detected oil-tea camellia fruit target with the corresponding target in the previous frame to form target tracking; managing and updating the target trajectory to ensure accurate maintenance of the movement trajectory of each oil-tea camellia fruit target; and finally constructing a rapid yield estimation model. The scope of application of this solution is relatively limited and is more suitable for yield assessment of small areas, local regions or specific sample trees. For the monitoring of large-scale oil-tea camellia planting areas, the workload and cost of image and video collection are relatively high, and it is difficult to achieve large-area and high-efficiency coverage. Summary of the Invention

[0005] To overcome the problem of poor detection accuracy in oil-tea camellia planting areas and the difficulty in achieving large-area and high-efficiency coverage in the prior art, the present invention provides a method and system for monitoring oil-tea camellia planting areas considering forest age and growth suitability.

[0006] The primary object of the present invention is to solve the above technical problems, and the technical solution of the present invention is as follows:

[0007] The first aspect of the present invention provides a method for monitoring oil-tea camellia planting areas considering forest age and growth suitability, including the following steps:

[0008] Extract the characteristics of newly planted oil-tea camellia forests and old oil-tea camellia forests from high-resolution images and multi-spectral images using deep learning models and machine learning models respectively, and output the distribution results of newly planted oil-tea camellia forests and old oil-tea camellia forests;

[0009] Obtain and utilize meteorological reanalysis data and topographic feature data, and combine with the suitability characteristics of oil-tea camellia growth to calculate the oil-tea camellia growth suitability scores in each dimension, and perform normalization processing on the scores to obtain an oil-tea camellia growth suitability distribution map;

[0010] Fuse the distribution results of newly planted oil-tea camellia forests and old oil-tea camellia forests to obtain a preliminary distribution map of the oil-tea camellia planting area;

[0011] Use the oil-tea camellia growth suitability distribution map as a prior knowledge reference, and impose suitability constraint conditions on the preliminary distribution map of the oil-tea camellia planting area to obtain a corrected distribution map of the oil-tea camellia planting area.

[0012] Further, the deep learning model adopts an encoder-decoder architecture. The encoder extracts multi-level features of the newly planted oil-tea camellia forest, and the multi-level features are transmitted to the decoder and converted into the final segmentation result. The machine learning model is a random forest model.

[0013] Further, the method for extracting the features of the newly planted oil-tea camellia forest and the features of the old oil-tea camellia forest includes the following steps:

[0014] Use high-resolution remote sensing images to train a deep learning model to extract the features of the newly planted oil-tea camellia forest, and apply the trained model to the area to be predicted to output the distribution result of the newly planted oil-tea camellia forest;

[0015] Use the distribution result of the newly planted oil-tea camellia forest as a mask, and train a machine learning model based on multi-spectral images to extract the features of the old oil-tea camellia forest, and apply the trained model to the area of the image to be predicted to output the distribution result of the old oil-tea camellia forest.

[0016] Further, using high-resolution remote sensing images to train a deep learning model to extract the features of the newly planted oil-tea camellia forest includes the following steps:

[0017] Use high-resolution remote sensing images and expert experience judgment, combined with on-site research, to visually interpret the area of the newly planted oil-tea camellia forest and perform refined annotation to obtain the vector label of the newly planted oil-tea camellia forest;

[0018] Rasterize the vector label of the newly planted oil-tea camellia forest to obtain a binary raster label;

[0019] Pair the high-resolution image with the raster label, and use non-overlapping sampling technology to obtain the sample data of the newly planted oil-tea camellia forest from the image, and divide the sample data into a training set and a validation set;

[0020] Use the training set and the validation set to train a deep learning model to extract the multi-level features of the newly planted oil-tea camellia forest;

[0021] Apply the trained deep learning model to the area to be predicted to output the distribution result of the newly planted oil-tea camellia forest.

[0022] Further, using the distribution result of the newly planted oil-tea camellia forest as a mask, and training a machine learning model based on multi-spectral images to extract the features of the old oil-tea camellia forest includes the following steps:

[0023] Obtain satellite images and perform data preprocessing to obtain multi-spectral images;

[0024] Calculate the multi-spectral image features, terrain features, polarization features, and temporal spectral features of the multi-spectral image, and fuse the features and add them to the image;

[0025] Within the study area, using the distribution results of the new oil-tea camellia forests as a mask, sample points of old oil-tea camellia forests are screened in the remaining area and randomly divided into a training set and a validation set;

[0026] Using the training set and the validation set to train a machine learning model to extract multi-level features of old oil-tea camellia forests;

[0027] Applying the trained machine learning model to the area to be predicted and outputting the distribution results of old oil-tea camellia forests.

[0028] Furthermore, the method for obtaining satellite images and performing data preprocessing is as follows: Using the GEE platform, calling Sentinel-2 or Landsat series satellite images, DEM data, and Sentinel-1 images within the same time range respectively, screening images for a specified period, removing clouds and selecting required bands, and performing radiometric correction, atmospheric correction, and geometric correction on the images to obtain clear cloud-free multi-spectral images.

[0029] Furthermore, the method for obtaining the oil-tea camellia growth suitability distribution map includes the following steps:

[0030] Collecting literature materials and survey data, organizing the meteorological and environmental conditions required for the growth of oil-tea camellia, constructing a suitability evaluation system for the growth of oil-tea camellia, and dividing it into n grades according to the suitability degree;

[0031] Obtaining meteorological reanalysis data and topographic feature data, performing reprojection, resampling, and data type conversion on the data, and combining the suitability evaluation system for the growth of oil-tea camellia to calculate multi-dimensional oil-tea camellia growth suitability scores;

[0032] Creating a score grid in the area to be predicted, calculating the pixel scores for each environmental factor, accumulating the scores of each factor, calculating the comprehensive suitability score for each pixel, and performing normalization processing on the comprehensive suitability score to obtain the oil-tea camellia growth suitability distribution map.

[0033] Furthermore, the meteorological reanalysis data and the topographic feature data are ERA5 reanalysis data and Copernicus topographic data respectively.

[0034] Furthermore, the method for correcting the oil-tea camellia planting area distribution map includes the following steps:

[0035] Initializing the suitability score threshold, creating a boolean layer on the area to be predicted to indicate whether the suitability score of the corresponding area pixel is higher than the threshold;

[0036] Retaining the pixels that were originally judged as oil-tea camellia land and whose suitability scores are higher than the threshold;

[0037] The statistical data of the introduced oil-tea camellia planting areas and areas are used as auxiliary verification to adjust the suitability scoring threshold, and the distribution of high-confidence oil-tea camellia planting areas is obtained.

[0038] In the second aspect of the present invention, an oil-tea camellia planting area monitoring system considering forest age and growth suitability is provided, including a memory and a processor. The memory includes an oil-tea camellia planting area monitoring method program considering forest age and growth suitability. When the oil-tea camellia planting area monitoring method program considering forest age and growth suitability is executed by the processor, the steps of an oil-tea camellia planting area monitoring method considering forest age and growth suitability are implemented.

[0039] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0040] By making full use of different types of remote sensing data and combining deep learning models, the present invention accurately extracts and corrects oil-tea camellia planting areas, has strong adaptability and flexibility, and is suitable for large-scale monitoring and management of oil-tea camellia resources. At the same time, this solution combines the growth suitability evaluation of oil-tea camellia, not only considers the distribution of oil-tea camellia planting areas, but also can further optimize the monitoring results, exclude unsuitable areas, and improve the accuracy. The new oil-tea camellia forests and old oil-tea camellia forests are extracted layer by layer and corrected in combination with growth suitability data to ensure that the monitoring results are not only accurate but also meet environmental requirements, effectively dealing with the problems of large inter-class differences and small intra-class differences existing in oil-tea camellias of different forest ages, and having strong generality and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to make the objectives and technical solutions of the present invention clearer, the following drawings are provided by the present invention and described:

[0042] Figure 1 It is the method flow chart provided by the embodiment of the present invention;

[0043] Figure 2 It is the overall technical roadmap provided by the embodiment of the present invention;

[0044] Figure 3 It is a sample example of the newly planted oil-tea camellia forest dataset provided by the embodiment of the present invention;

[0045] Figure 4 It is a schematic diagram of the correction process of the oil-tea camellia forest extraction result provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] In order to be able to more clearly understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments may be combined with each other.

[0047] In the following description, many specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited by the specific embodiments disclosed below.

[0048] Embodiment 1:

[0049] The present invention provides a method for monitoring oil-tea camellia planting areas considering forest age and growth suitability. In order to obtain a dataset of newly planted oil-tea camellia forests from high-resolution remote sensing images, 50 ESRI Word Imagery Wayback images located in various regions of Guangdong Province were downloaded. The resolution of the images is approximately 0.5 meters, and all are clear cloud-free images obtained through manual inspection. Figure 1 The following shows a flowchart of a method for monitoring oil-tea camellia planting areas considering forest age and growth suitability provided by the present invention. Figure 2 The following shows the overall technical roadmap of a method for monitoring oil-tea camellia planting areas considering forest age and growth suitability. The specific steps are as follows:

[0050] S1: Use a deep learning model and a machine learning model respectively to extract the features of newly planted oil-tea camellia forests and the features of old oil-tea camellia forests from high-resolution images and multi-spectral images, and output the distribution results of newly planted oil-tea camellia forests and the distribution results of old oil-tea camellia forests.

[0051] More specifically, the deep learning model adopts an encoder-decoder architecture. The encoder extracts multi-level features of newly planted oil-tea camellia forests, and the multi-level features are transmitted to the decoder and converted into the final segmentation result; the machine learning model is a random forest model.

[0052] More specifically, the method for extracting the features of newly planted oil-tea camellia forests and the features of old oil-tea camellia forests includes the following steps:

[0053] Use high-resolution remote sensing images to train a deep learning model to extract the features of newly planted oil-tea camellia forests, and apply the trained model to the area to be predicted to output the distribution result of newly planted oil-tea camellia forests.

[0054] The specific process is as follows:

[0055] First, refer to the approximate distribution range of the oil-tea camellia planting area to delimit the sample collection area. Subsequently, download the corresponding high-resolution images from the open-source ESRI Word Imagery website (https: / / livingatlas.arcgis.com / wayback). According to the data description document, perform preprocessing operations such as radiometric calibration and histogram matching on the downloaded image data, and screen out clear cloud-free images from them to improve the efficiency of subsequent annotation.

[0056] Secondly, 50 selected high-resolution ArcGIS images were visually interpreted by experts to label the newly planted camellia oleifera forest areas on the images. For uncertain areas, verification was carried out by combining multi-source data comparison and field research, and then the vector labels of the newly planted camellia oleifera forest corresponding to each high-resolution image were obtained.

[0057] Finally, all the labeled vector data were converted into binary class label images, where camellia oleifera and other classes were represented by the values 255 and 0 respectively. A certain number of samples were collected from the sample area images and labels in a non-overlapping random sampling manner. The paired image and label data were cropped into 512×512 sample blocks to meet the requirements of model training on the GPU, obtaining a monitoring dataset of newly planted camellia oleifera forest under high-resolution images, and dividing the sample data into a training set and a validation set;

[0058] In this embodiment, through the above steps, a total of 4265 pairs of newly planted camellia oleifera forest samples were obtained, and some sample examples are as Figure 3 shown. These samples were randomly divided into a training set and a validation set in a ratio of 3:1.

[0059] The training set and the validation set were used to train a deep learning model to extract multi-level features of the newly planted camellia oleifera forest;

[0060] In order to extract the newly planted camellia oleifera forest from high-resolution images, this study constructed a deep learning-based extraction network for the newly planted camellia oleifera forest. In this example, the model consists of three main modules: (1) a pixel-level feature extraction module based on SAM, which is used to extract pixel-level features of the image; (2) an instance-level feature extraction module based on Transformer, which is used to learn the target features in the image; (3) a classifier, which is used to finally extract the distribution of camellia oleifera in the image.

[0061] Specifically, the pixel-level feature extraction module introduces the SegmentAnything Model (SAM) as the backbone network, making full use of SAM's powerful edge feature extraction ability to extract pixel-level features of newly planted oil-tea camellia forests from high-resolution images. These features are passed to a pixel decoder and, through progressive upsampling, generate embedding vectors for each pixel. These embedding vectors contain local and global context information for each pixel in the image, providing a basis for subsequent mask prediction. Next, the instance-level feature extraction module receives the pixel features from the previous step, assigns importance weights to different parts of the features through the attention mechanism, and generates several embedding vectors. These embedding vectors represent the features of different regions in the image, and each embedding vector independently generates a class prediction and a corresponding mask embedding vector, that is, the potential targets to be extracted in the image. Finally, the classifier will receive the aforementioned features and convert them into the final segmentation result. By performing a dot product operation on the embedding vectors obtained from the pixel-level feature extraction module and the mask embedding vectors of the instance-level feature extraction module, multiple potentially overlapping binary masks are predicted. These masks are processed through the sigmoid activation function to determine whether each pixel belongs to the newly planted oil-tea camellia forest class.

[0062] After constructing the network structure, appropriate parameters are set for model training. In this embodiment, the number of training epochs is set to 200, the optimizer is Adam, the initial learning rate is 0.0001, and the training batch size is 8. The deep learning model is trained using the aforementioned constructed training set, and at the same time, the model performance is continuously examined in real time through the validation set, and the model parameters with the highest current accuracy are continuously saved.

[0063] This embodiment takes Xingning City, Guangdong Province as the research area. Finally, the proposed model trained based on the newly planted oil-tea camellia forest dataset of high-resolution remote sensing images is applied to the high-resolution images of Xingning City to output the distribution results of the newly planted oil-tea camellia forests.

[0064] Using the distribution results of the newly planted oil-tea camellia forests as a mask, a machine learning model is trained based on multi-spectral images to extract the features of the old oil-tea camellia forests, and the trained model is applied to the area of the image to be predicted to output the distribution results of the old oil-tea camellia forests.

[0065] The specific process is as follows:

[0066] First, the required data is preprocessed on the GEE platform. The Sentinel-2 images within the specific time range from January 1, 2023 to December 31, 2023 are selected using the correlation function. The cloud layer in the image is removed by the built-in declouding function, and key bands such as B2-B8, B8A, B11, and B12 are selected; then, the band information of the image is used to calculate a variety of vegetation indices, such as the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Building Index (BSI), etc.; at the same time, with the help of the terrain analysis tools provided by the GEE platform, terrain features such as altitude, slope, and aspect are calculated, and these newly calculated indices and features are added to the image as new bands. Sentinel-1 images within the same time range are obtained, and statistical features such as mean, variance, correlation, and their combined features are calculated for their VV and VH polarization modes.

[0067] In synthetic aperture radar (SAR) imaging, polarization mode refers to the polarization mode used by the radar when transmitting and receiving electromagnetic waves. For the common dual-polarization (Dual-Pol) mode, such as Sentinel-1 data, it generally includes the following two common polarization combinations: VV / VH or HH / HV. The "V" here stands for vertical polarization, and the "H" stands for horizontal polarization; the first letter stands for transmit polarization, and the second letter stands for receive polarization.

[0068] Next, the time series spectral features are combined. In terms of the combination of time series spectral features, first of all, the statistical characteristics of the NDVI and BSI indices in different time periods (such as monthly and quarterly) are calculated for the time series features, including the mean, median, maximum, minimum, phase and amplitude. For example, when calculating the monthly mean NDVI, the image data of each month is looped through, and the reduceRegion function of the GEE platform is used to calculate the monthly NDVI mean in the form of regional averages; then the spectral characteristics of the Sentinel-2 image, the newly calculated vegetation index and terrain characteristics, the polarization pattern statistical characteristics of the Sentinel-1 image, and the time series characteristics are combined into a feature set. The feature set ultimately contains 50 different features, providing rich data information for subsequent classification.

[0069] Finally, the random forest algorithm is used for classification. In the study area, using the distribution result of the new oil-tea camellia forest as a mask, sample points of the old oil-tea camellia forest are screened in the remaining area and randomly divided into a training set and a validation set. Sample areas are extracted from the feature set, and a total of 2,101 sample points are obtained, among which 1,471 are used as the training set and 630 are used as the validation set. Taking the random forest algorithm as the classifier, the number of decision trees is set to 100, and the number of features considered when each node is split is sqrt(50), that is, about 7 features. Through multiple experiments, it is determined that this parameter setting can achieve good classification results on this data set. Based on the training set data, a classifier is trained, and using the classification function of the GEE platform, the trained classifier is applied to the study area image to generate the classification result of the old oil-tea camellia forest.

[0070] In the model selection of the present invention, a deep learning model is adopted for the newly planted oil-tea camellia forest, while a random forest model is selected for the old oil-tea camellia forest, mainly based on the different characteristics presented by the two in remote sensing images. Since the newly planted oil-tea camellia forest has not yet formed a dense canopy and the row spacing is relatively obvious, clearer texture differences are more likely to appear in sub-meter high-resolution images, and the deep learning model is good at capturing such microscopic features. On the contrary, the visual difference between the old oil-tea camellia forest and other forest lands in high-resolution images is not significant. Using multi-spectral images can better utilize the rich spectral information to distinguish the old oil-tea camellia forest from other forest lands, and the random forest algorithm is widely used and has stable effects in multi-spectral classification. To cope with the significant differences in characteristics between the newly planted and old oil-tea camellia forests, the solution adopts the idea of hierarchical extraction: deep learning is selected for the texture features of high-resolution images, and random forest is selected for the spectral differences of multi-spectral images.

[0071] S2: Obtain and utilize meteorological reanalysis data and terrain feature data, combine with the suitability characteristics of oil-tea camellia growth, calculate the oil-tea camellia growth suitability scores for each dimension, and perform normalization processing on the scores to obtain the oil-tea camellia growth suitability distribution map.

[0072] More specifically, the meteorological reanalysis data and terrain feature data are ERA5 reanalysis data and Copernicus terrain data respectively.

[0073] In this embodiment, Xingning City, Guangdong Province is used as the study area to carry out the assessment of oil-tea camellia growth suitability based on meteorological data, aiming to clarify the suitability distribution of oil-tea camellia growth in this area.

[0074] In terms of data collection and preprocessing. The meteorological data are precipitation data from August 1, 2023 to September 30, 2023 obtained from the NASA / GPM_L3 / IMERG_V06 dataset on the GEE platform and temperature data from October 1, 2023 to December 31, 2023, January 2023, and July 2023 obtained from the ECMWF / ERA5_LAND / MONTHLY_AGGR dataset. After screening, mean calculation, reprojection (EPSG:4326, resolution 1km), and bilinear resampling (resolution 500m), single-band precipitation data (precipitation) and temperature data for different time periods (temperature, januaryTemp, and julyTemp) are obtained, and the data types are checked. If they do not meet the requirements, they are converted to the Int type.

[0075] The soil data are soil data such as organic matter content (organicMatter), total nitrogen content (totalNitrogen), and pH value (pH) obtained from the China Soil Organic Matter Dataset of the National Tibetan Plateau Data Center of Sciences. They are reprojected, resampled, and data type checked to ensure that the data resolution is 500m, and selected and renamed according to the actual band names.

[0076] The terrain data are DEM data obtained from the USGS / SRTMGL1_003 dataset. After reprojection, resampling, and data type checking, slope (slope) and aspect (aspect) data are calculated.

[0077] In terms of constructing the evaluation system for the suitability of Camellia oleifera growth, by widely collecting literature materials and field research data, the requirements of Camellia oleifera growth for meteorological, environmental and other conditions in different phenological periods are clarified, and the suitability degree of Camellia oleifera growth is divided into three grades, namely suitable (score 2 points), relatively suitable (score 1 point), and unsuitable (score 0 points).

[0078] When calculating the suitability scores of Camellia oleifera growth in each dimension, a suitability score (suitabilityScore) function is defined, and a single-band image (scoreImage) with all values of 0 is created to store the final scores. For each environmental factor, according to the suitability grades of Camellia oleifera growth, conditional judgment statements are used to calculate the score raster and accumulate it into scoreImage. The suitabilityScore function is called to obtain the suitability score raster at the pixel level.

[0079] S3: Integrate the distribution results of newly created oil-tea camellia forests and the distribution results of old oil-tea camellia forests to obtain a preliminary distribution map of the oil-tea camellia planting area; use the oil-tea camellia growth suitability distribution map as a prior knowledge reference, and impose suitability constraint conditions on the preliminary distribution map of the oil-tea camellia planting area to obtain a corrected distribution map of the oil-tea camellia planting area.

[0080] First, integrate the results of newly and old oil-tea camellia forests extraction to obtain the preliminary oil-tea camellia distribution result; then, determine the target size, resample the suitability score map to the size of the oil-tea camellia planting distribution map, normalize the resampled suitability score map, and use the oil-tea camellia growth suitability distribution map as a prior knowledge reference to correct the oil-tea camellia planting area distribution map.

[0081] The specific process is as follows:

[0082] Initialize the suitability score threshold, and create a boolean layer on the area to be predicted to indicate whether the suitability score of the corresponding area pixel is higher than the threshold;

[0083] Retain the pixels that were originally judged as oil-tea camellia land and whose suitability score is higher than the threshold;

[0084] Introduce the oil-tea camellia planting area and area statistical data as auxiliary verification to adjust the suitability score threshold to obtain a high-confidence oil-tea camellia planting area distribution. In this embodiment, this process is as Figure 4 shown. Taking Xingning City as an example, from left to right in the figure are: (a) The initial result of oil-tea camellia forest extraction; (b) The growth suitability distribution map; (c) The corrected result.

[0085] The present invention makes full use of different types of remote sensing data, combines deep learning models for accurate extraction and correction of the oil-tea camellia planting area, has strong adaptability and flexibility, and is suitable for large-scale monitoring and management of oil-tea camellia resources; at the same time, this solution combines the growth suitability assessment of oil-tea camellia, not only considers the distribution of the oil-tea camellia planting area, but also can further optimize the monitoring results, exclude unsuitable areas, and improve the accuracy; extract new and old oil-tea camellia forests in layers and correct them in combination with growth suitability data to ensure that the monitoring results are not only accurate but also meet environmental requirements, effectively handle the problem of large inter-class differences and small intra-class differences existing in oil-tea camellia of different forest ages, and have strong generality and stability.

[0086] Embodiment 2:

[0087] This embodiment provides an oil-tea camellia planting area monitoring system considering forest age and growth suitability, including a memory and a processor. The memory includes an oil-tea camellia planting area monitoring method program considering forest age and growth suitability. When the oil-tea camellia planting area monitoring method program considering forest age and growth suitability is executed by the processor, it realizes the steps of an oil-tea camellia planting area monitoring method as described in Embodiment 1.

[0088] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the claims of the present invention.

Claims

1. A method for monitoring tea oil planting areas taking into account forest age and growth suitability, characterized in that: The steps include: The deep learning model and machine learning model are used to extract the characteristics of the newly planted oil-tea forest and the characteristics of the old oil-tea forest from the high-resolution images and multispectral images respectively, and the distribution results of the newly planted oil-tea forest and the distribution results of the old oil-tea forest are output; Obtain and use meteorological reanalysis data and terrain feature data, combine with the suitability characteristics of oil-tea growth, calculate the suitability scores of oil-tea growth in each dimension, normalize the scores, and obtain a distribution map of suitability for oil-tea growth; The distribution results of newly planted tea oil forests were integrated with those of old tea oil forests to obtain a preliminary distribution map of tea oil planting areas. Using the tea oil growth suitability distribution map as a priori knowledge reference, suitability constraints were imposed on the preliminary distribution map of tea oil planting areas to obtain a revised tea oil planting area distribution map.

2. A method for monitoring tea-oil planting areas taking into account forest age and growth suitability according to claim 1, characterized in that: The deep learning model adopts an encoder-decoder architecture. The encoder extracts multi-level features of the newly planted oil tea forest. The multi-level features are passed to the decoder and converted into final segmentation results. The machine learning model is a random forest model.

3. The method for monitoring the oil-tea camellia planting area taking into account the forest age and growth suitability according to claim 1, characterized in that: The method for extracting the characteristics of a newly planted oil-tea tree forest and the characteristics of an old oil-tea tree forest comprises the following steps: Use high-resolution remote sensing images to train a deep learning model to extract the characteristics of newly planted tea oil forests, and apply the trained model to the area to be predicted to output the distribution results of newly planted tea oil forests; Using the distribution results of newly planted oil tea forests as a mask, a machine learning model is trained based on multispectral images to extract the features of old oil tea forests. The trained model is then applied to the image area to be predicted to output the distribution results of old oil tea forests.

4. A method for monitoring tea-oil planting areas taking into account forest age and growth suitability according to claim 3, characterized in that: Using high-resolution remote sensing images to train a deep learning model to extract the characteristics of newly planted tea oil forests includes the following steps: Using high-resolution remote sensing images and expert experience, combined with field research, visual interpretation of the newly planted oil-tea tree forest area was carried out, and fine annotation was performed to obtain vector labels for the newly planted oil-tea tree forest; The vector labels of the newly planted oil-tea forest are rasterized to obtain binary raster labels; Pair high-resolution images with raster labels, use non-overlapping sampling technology to obtain new oil-tea forest sample data from the images, and divide the sample data into training sets and validation sets. Using the training set and the validation set to train a deep learning model to extract multi-level features of the newly planted tea oil forest; The trained deep learning model is applied to the area to be predicted, and the distribution results of newly planted tea oil forests are output.

5. The method for monitoring the oil-tea camellia planting area taking into account the forest age and growth suitability according to claim 3, characterized in that: Using the distribution results of the newly planted tea oil forest as a mask, the machine learning model was trained based on multispectral images to extract the characteristics of the old tea oil forest, including the following steps: Acquire satellite images and perform data preprocessing to obtain multispectral images; Calculate multispectral image features, terrain features, polarization features, and time-series spectral features using multispectral images, and fuse and add the features to the images; In the study area, the distribution results of new oil-tea forests were used as masks to select old oil-tea forest sample points in the remaining area and randomly divided them into training set and validation set; Using the training set and the validation set to train a machine learning model to extract multi-level features of the old tea oil forest; The trained machine learning model is applied to the area to be predicted and the distribution results of old tea oil forests are output.

6. A method for monitoring tea-oil planting areas taking into account forest age and growth suitability according to claim 5, characterized in that: The method for acquiring satellite images and performing data preprocessing is as follows: using the GEE platform, respectively calling Sentinel-2 or Landsat series satellite images, as well as DEM data and Sentinel-1 images within the same time range, screening images of a specified period, removing clouds and selecting the required bands, and performing radiation correction, atmospheric correction, and geometric correction on the images to obtain clear, cloud-free multispectral images.

7. A method for monitoring tea-oil planting areas taking into account forest age and growth suitability according to claim 1, characterized in that: The method for obtaining a distribution map of oil-tea camellia growth suitability comprises the following steps: Collect literature and survey data, sort out the meteorological and environmental conditions required for the growth of oil-tea camellia, build a suitability evaluation system for the growth of oil-tea camellia, and divide it into n levels according to the degree of suitability; Acquire meteorological reanalysis data and terrain feature data, reproject, resample and convert the data types of the data, and combine the suitable evaluation system for oil-tea growth to calculate the multi-dimensional suitability score for oil-tea growth; A score grid is created in the area to be predicted, and the pixel score is calculated for each environmental factor. The scores of each factor are accumulated to calculate the comprehensive suitability score of each pixel. The comprehensive suitability score is normalized to obtain a distribution map of oil-tea growth suitability.

8. A method for monitoring tea-oil planting areas taking into account forest age and growth suitability according to claim 7, characterized in that: The meteorological reanalysis data and terrain feature data are ERA5 reanalysis data and Copernicus terrain data respectively.

9. The method for monitoring tea-oil planting areas taking into account forest age and growth suitability according to claim 1, characterized in that: The method for correcting the distribution map of oil-tea camellia planting areas comprises the following steps: Initialize the suitability score threshold and create a Boolean value layer on the area to be predicted to indicate whether the suitability score of the corresponding area pixel is higher than the threshold; Pixels that were originally judged to be oil-tea land and whose suitability scores were higher than the threshold were retained; The oil-tea camellia planting regions and area statistics data were introduced as auxiliary verification to adjust the suitability score threshold and obtain the distribution of oil-tea camellia planting areas with high confidence.

10. A monitoring system for oil-tea camellia planting areas taking into account forest age and growth suitability, characterized in that: The system includes: a memory and a processor, wherein the memory includes a program for monitoring a tea oil planting area taking into account forest age and growth suitability, and when the program for monitoring a tea oil planting area taking into account forest age and growth suitability is executed by the processor, the steps of a method for monitoring a tea oil planting area taking into account forest age and growth suitability as described in any one of claims 1 to 9 are implemented.

Images