Data labeling method and system applied to high-cold steep slope plant screening

Abstract

The application provides a data labeling method and system applied to high-cold steep slope plant screening, and machine learning technology is used to realize automatic labeling of data. Through multi-source data collaborative collection of the high-cold steep slope region, topographic structure data, plant image data and environment monitoring data of the high-cold steep slope region are synchronously acquired, and a time-space correlation data set is constructed; a mapping model of vegetation index and cold-resistant gene expression product is constructed based on a transfer learning framework, vegetation index features are extracted from the plant image data, and the mapping model is used to convert the vegetation index features into corresponding cold-resistant gene expression product features; a preset slope compensation model is called to correct topographic distortion of the topographic structure data, and standardized topographic data is obtained, and the cold-resistant gene expression product features are combined to perform spatial feature calibration processing; time sequence trend analysis is performed according to the cold-resistant gene expression product features and the environment monitoring data, and a plant growth adaptability label is generated. The application can realize efficient and accurate labeling.

Description

Technical Field

[0001] This invention relates to the field of data processing, and more specifically, to a data labeling method and system for screening plants on alpine and steep slopes. Background Technology

[0002] In ecological environment governance and vegetation restoration projects, plant selection in high-altitude, cold, and steep slope areas is a crucial step. Accurate assessment of plant stress resistance is essential to identify plant varieties adapted to this unique habitat. Currently, plant stress resistance data is typically obtained through manual collection of plant samples combined with laboratory analysis, or by using sensors to collect and label plant phenotypic features. During data processing, raw topographic data is often used for feature extraction, and the labeling results are usually presented as static labels. However, in high-altitude, cold, and steep slope areas, manual collection and laboratory analysis are hampered by terrain conditions, making them difficult to implement and unsuitable for large-scale plant selection. There is a lack of direct correlation between sensor-collected phenotypic features and latent stress resistance traits in plants. Directly using raw topographic data can lead to feature extraction biases due to the complex variations in steep slope terrain. Static labels fail to reflect changes in plant stress resistance at different growth stages. These factors collectively limit the applicability and accuracy of existing data labeling methods in the special context of high-altitude, cold, and steep slopes, making it difficult to effectively support precise plant selection. Summary of the Invention

[0003] This invention provides a data annotation method and system for screening plants on alpine and steep slopes.

[0004] In a first aspect, embodiments of the present invention provide a data annotation method for screening plants on alpine and steep slopes. The method includes: conducting multi-source collaborative data collection on alpine and steep slope areas, simultaneously acquiring topographic structure data, plant image data, and environmental monitoring data of the alpine and steep slope areas, and constructing a spatiotemporal associated dataset through timestamp alignment; constructing a mapping model between vegetation indices and cold-resistant gene expression products based on a transfer learning framework, extracting vegetation index features from plant image data in the spatiotemporal associated dataset, and converting the vegetation index features into corresponding cold-resistant gene expression product features through the mapping model; calling a preset slope compensation model to perform terrain distortion correction processing on the topographic structure data in the spatiotemporal associated dataset to obtain standardized topographic data that eliminates the influence of slope, and performing spatial feature calibration processing on the cold-resistant gene expression product features in combination with the standardized topographic data; performing time-series trend analysis based on the cold-resistant gene expression product features after spatial feature calibration processing and the environmental monitoring data in the spatiotemporal associated dataset, calculating growth adaptability indicators for different time periods through a plant growth adaptability assessment algorithm, and generating plant growth adaptability labels based on the growth adaptability indicators.

[0005] Secondly, embodiments of the present invention provide a computer system, comprising: a memory storing a computer program; and a processor for loading the computer program to implement the data labeling method for screening plants on alpine and steep slopes as described above.

[0006] The data annotation method for plant screening on alpine and steep slopes provided by this invention synchronously acquires topographic structure data, plant image data, and environmental monitoring data through multi-source collaborative data collection in alpine and steep slope areas, and constructs a spatiotemporally correlated dataset. This overcomes the limitations of traditional single-source data collection modes that lack spatiotemporal correlation between data, enabling topographic, plant, and environmental data to form a holistic data volume with inherent relationships, laying a data foundation for the accurate extraction of latent stress resistance traits. Based on a transfer learning framework, a mapping model between vegetation indices and cold-resistance gene expression products is constructed, converting vegetation index features in plant image data into cold-resistance gene expression product features. This effectively solves the problems of high cost and long cycle caused by the reliance on molecular biology experiments in traditional latent stress resistance trait annotation, realizing the transformation from observable phenotypic features to latent biological characteristics. This method employs cross-modal transformation of plant characteristics; it utilizes a slope compensation model to correct terrain distortion in terrain structure data and combines standardized terrain data to calibrate the spatial features of latent stress resistance traits. This effectively eliminates the interference of steep, cold-climate terrain tilt on plant feature extraction, improving the spatial accuracy of latent stress resistance traits. Based on the calibrated latent stress resistance traits and environmental monitoring data, it performs time-series trend analysis to generate plant growth adaptability labels. This overcomes the limitation of traditional static labels, which cannot reflect the changes in plant stress resistance over time, enabling the labeling results to present the plant's growth adaptability status at different times. Through the organic combination of the above technical features, this method can achieve efficient and accurate labeling of latent stress resistance traits in plants in the special scenario of steep, cold-climate slopes, providing reliable data support for the screening of plants on steep, cold-climate slopes. Attached Figure Description

[0007] Figure 1 This is a flowchart of a data annotation method for screening plants on steep, cold slopes, provided by an embodiment of the present invention.

[0008] Figure 2 This is a schematic diagram of the composition of a computer system provided in an embodiment of the present invention. Detailed Implementation

[0009] Please see Figure 1 , Figure 1 A flowchart illustrating a data annotation method for screening plants on steep, cold slopes, provided in this embodiment of the invention, is shown. This method can be executed by a computer system and may include the following steps: Step S100: Conduct multi-source collaborative data collection in the high-altitude and steep slope area, simultaneously acquire topographic structure data, plant image data and environmental monitoring data of the high-altitude and steep slope area, and construct a spatiotemporal related dataset by aligning the timestamps.

[0010] Topographic structure data describes the topographic features of high-altitude, cold, and steep slope regions, such as elevation, slope, and aspect, reflecting the topographic relief and landform of the area. Plant image data provides visual information about plants in these regions, including their appearance, morphology, and distribution. Environmental monitoring data is obtained by monitoring environmental factors in high-altitude, cold, and steep slope regions, covering information such as temperature, humidity, light intensity, and soil moisture, reflecting the environmental conditions of the area.

[0011] For terrain structure data, LiDAR technology can be used to acquire high-precision 3D terrain data for high-altitude, cold, and steep slope areas, and the geographical location information of the terrain data can be determined using the Global Positioning System (GPS). For plant image data, drones equipped with high-resolution cameras can be used to conduct aerial photography of high-altitude, cold, and steep slope areas to acquire plant image data. For environmental monitoring data, multiple environmental monitoring stations can be set up in high-altitude, cold, and steep slope areas, equipped with devices such as temperature sensors, humidity sensors, and light sensors to collect environmental data in real time.

[0012] Timestamp alignment unifies the collection times of different types of data, ensuring data consistency over time. By adding timestamp information to each data point and then sorting and matching the data chronologically, a spatiotemporally correlated dataset can be constructed. For example, topographic structure data, plant image data, and environmental monitoring data can be sorted according to their collection times, linking different types of data from the same point in time together to form a dataset containing diverse data information.

[0013] Step S200: Construct a mapping model between vegetation indices and cold-resistant gene expression products based on a transfer learning framework. Extract vegetation index features from plant image data in a spatiotemporally correlated dataset, and convert the vegetation index features into corresponding cold-resistant gene expression product features through the mapping model.

[0014] In this embodiment of the invention, the transfer learning framework can apply models and knowledge trained on other related fields or datasets to the task of screening plants on steep, cold slopes, improving the training efficiency and performance of the model. The vegetation index is a numerical value calculated from the spectral information in plant image data, reflecting the plant's growth status, health, and photosynthetic capacity. Cold-resistance gene expression products are substances produced by plants during the cold-resistance process, such as proteins and enzymes; the expression levels of these substances are closely related to the plant's cold-resistance.

[0015] The purpose of constructing a mapping model between vegetation indices and cold-resistance gene expression products is to establish a mathematical relationship between the two, so as to predict the characteristics of cold-resistance gene expression products through vegetation index features. Neural network models, such as multilayer perceptrons (MLPs), can be used to construct this mapping model. A multilayer perceptron consists of an input layer, hidden layers, and an output layer. The input layer receives vegetation index features, the hidden layers perform nonlinear transformations and feature extraction on the input features, and the output layer outputs the corresponding cold-resistance gene expression product features.

[0016] Extracting vegetation index features from plant image data in a spatiotemporally correlated dataset requires processing and analysis of the plant image data. Spectral analysis techniques can be used to separate and calculate the spectral bands in the plant image data, yielding different vegetation indices, such as the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI). These vegetation indices are then input into a mapping model as vegetation index features. Through model calculation and prediction, the vegetation index features are converted into corresponding cold-hardy gene expression product features. This achieves the conversion from plant image data to cold-hardy gene expression product features, providing an important basis for subsequent screening of cold-hardy plants.

[0017] In one implementation, step S200 may specifically include the following steps S210-S240: Step S210: Perform spectral band separation processing on the plant image data in the spatiotemporal correlation dataset, extract near-infrared band reflectance data and visible light band reflectance data, and generate initial vegetation index features through a band combination algorithm.

[0018] Spectral band separation processing involves separating different spectral bands in plant image data to extract near-infrared reflectance data and visible light reflectance data separately. Near-infrared reflectance data represents the reflectance of plants in the near-infrared spectral band and is closely related to chlorophyll content and leaf structure. Visible light reflectance data represents the reflectance of plants in the visible light spectral band, reflecting their color and appearance characteristics. Band combination algorithms combine near-infrared and visible light reflectance data to generate initial vegetation index features. A possible band combination algorithm is the Normalized Difference Vegetation Index (NDVI) algorithm, whose formula is: NDVI = (Near-infrared reflectance - Visible light reflectance) / (Near-infrared reflectance + Visible light reflectance). This algorithm combines near-infrared and visible light reflectance data into a single value, the Normalized Difference Vegetation Index, which serves as the initial vegetation index feature.

[0019] When performing spectral band separation, image processing techniques such as filtering and segmentation can be used to separate different spectral bands from plant image data. For example, a bandpass filter can be used to retain only information from the near-infrared or visible light bands, thereby extracting near-infrared and visible light reflectance data. Then, according to the formula of the band combination algorithm, the extracted reflectance data is substituted into the formula for calculation to obtain the initial vegetation index features.

[0020] In one implementation, step S210 may specifically include the following steps S211-S215: Step S211: Perform radiometric correction on the plant image data to eliminate the influence of atmospheric scattering and sensor response differences on spectral reflectance, and generate a standardized spectral reflectance dataset.

[0021] The purpose of radiometric correction is to eliminate the effects of atmospheric scattering and sensor response differences on spectral reflectance. Atmospheric scattering causes light to scatter during propagation, resulting in inaccurate received spectral reflectance data. Sensor response differences refer to the potential for variations in the response of different sensors to the same target, which also affects the accuracy of spectral reflectance data.

[0022] The standardized spectral reflectance dataset is a dataset obtained after radiometric correction. The spectral reflectance data in it has eliminated the influence of atmospheric scattering and sensor response differences, and has higher accuracy and comparability.

[0023] When performing radiometric correction, physical model-based methods, such as the Atmospheric Radiative Transfer Model (MODTRAN), can be employed. This model simulates the transmission of light in the atmosphere, calculates the impact of atmospheric scattering on spectral reflectance, and performs corrections. Simultaneously, the sensor response can be calibrated by measuring a standard target with known reflectance to obtain the sensor's response curve, which is then used to correct the plant image data. Through these processes, the spectral reflectance data in the plant image data is converted into standardized spectral reflectance data, generating a standardized spectral reflectance dataset.

[0024] Step S212: Perform band separation operation on the standardized spectral reflectance dataset according to the preset band division threshold to separate the reflectance data sequence in the near-infrared band and the reflectance data sequence in the visible light band.

[0025] The preset band division threshold is a value pre-set based on the characteristics of the spectral bands and research needs, used to distinguish between the near-infrared band and the visible light band. The near-infrared band typically refers to the spectral range with wavelengths between 700 and 2500 nanometers, while the visible light band refers to the spectral range with wavelengths between 380 and 700 nanometers.

[0026] Band separation is the process of separating spectral data in a standardized spectral reflectance dataset according to a preset band division threshold, resulting in reflectance data sequences for the near-infrared band and the visible light band. This can be achieved by iterating through each spectral data point in the standardized spectral reflectance dataset and classifying it into either the near-infrared or visible light band based on its wavelength and the preset band division threshold. For example, if a spectral data point has a wavelength greater than 700 nanometers, it is classified as reflectance data in the near-infrared band; if the wavelength is less than 700 nanometers, it is classified as reflectance data in the visible light band.

[0027] By performing band separation, the standardized spectral reflectance dataset is decomposed into reflectance data sequences in the near-infrared band and the visible light band, providing essential foundational data for subsequent band combination and vegetation index calculations. These two reflectance data sequences can respectively reflect the reflectance characteristics of plants in the near-infrared and visible light bands, helping to more accurately analyze plant growth status and health.

[0028] Step S213: Perform pixel-by-pixel band operation processing on the near-infrared band reflectance data sequence and the visible light band reflectance data sequence, and calculate the initial vegetation index characteristics through a preset band combination algorithm.

[0029] Pixel-by-pixel band operations involve performing corresponding calculations on each pixel in the near-infrared and visible light reflectance data sequences. The preset band combination algorithm is a pre-defined algorithm based on research requirements and the definition of vegetation indices. It is used to combine near-infrared and visible light reflectance data into initial vegetation index features, such as the Normalized Difference Vegetation Index (NDVI) algorithm described earlier.

[0030] By performing pixel-by-pixel band operations on all pixels, a series of normalized vegetation index values ​​are obtained, which constitute the initial vegetation index features.

[0031] Step S214: Spatial smoothing is performed on the initial vegetation index features, and the median filtering algorithm is used to eliminate the interference of salt-and-pepper noise on the feature values, generating the noise-suppressed initial vegetation index features.

[0032] Spatial smoothing aims to eliminate noise in an image, making feature values ​​smoother and more continuous. Median filtering algorithms sort the values ​​of each pixel and its neighboring pixels in the image, then take the median value as the new value for that pixel. When performing median filtering on the initial vegetation index features, a neighborhood window, such as a 3×3 or 5×5 window, is first determined. For each pixel in the initial vegetation index features, the values ​​of all pixels within its neighborhood window are sorted, and the median value is taken as the new value for that pixel. Through median filtering, the interference of salt-and-pepper noise on the initial vegetation index feature values ​​can be effectively eliminated, generating noise-suppressed initial vegetation index features. The noise-suppressed initial vegetation index features are smoother and more accurate.

[0033] Step S215: Adjust the dimensions of the initial vegetation index features after noise suppression to a format that matches the input dimensions of the migration feature vector.

[0034] The transfer feature vector is a feature vector used in the transfer learning framework and has input dimension requirements. The dimension of the initial vegetation index features after noise suppression may not match the input dimension of the transfer feature vector, so dimension adjustment is required.

[0035] The method for dimensionality adjustment can be chosen based on the input dimensionality requirements of the transfer feature vector. For example, if the input dimension of the transfer feature vector is a one-dimensional vector, and the initial vegetation index feature after noise suppression is two-dimensional image data, the two-dimensional image data can be flattened and converted into a one-dimensional vector. For example, the values ​​of each pixel in the two-dimensional image data can be arranged into a one-dimensional vector in row-major or column-major order.

[0036] If the input dimension of the transfer feature vector is a three-dimensional tensor, while the noise-suppressed initial vegetation index feature is a one-dimensional vector, the one-dimensional vector can be expanded and reshaped to convert it into a three-dimensional tensor. For example, additional dimensions can be added as needed, and the size of each dimension can be adjusted to match the input dimension of the transfer feature vector. By adjusting the dimensions of the noise-suppressed initial vegetation index feature to match the input dimension of the transfer feature vector, it can be ensured that the initial vegetation index feature can be correctly input into the transfer learning framework.

[0037] Step S220: Input the initial vegetation index features into the pre-trained feature transfer network, and perform high-dimensional feature space mapping processing on the initial vegetation index features through the domain adaptation layer of the feature transfer network to generate transfer feature vectors.

[0038] Pre-trained feature transfer networks are network models pre-trained on other relevant datasets, possessing certain feature extraction and transfer capabilities. The domain adaptation layer of a feature transfer network addresses the data distribution differences between the source and target domains, enabling cross-domain feature transfer. High-dimensional feature space mapping is the process of mapping initial vegetation index features from a low-dimensional space to a high-dimensional space. In high-dimensional space, the relationships between features are more complex and richer, better representing the intrinsic characteristics of the data. By using the domain adaptation layer of a feature transfer network to perform high-dimensional feature space mapping on the initial vegetation index features, the initial vegetation index features can have better discriminative power and transferability in high-dimensional space.

[0039] Domain adaptation layers can employ adversarial training to achieve cross-domain feature transfer. For example, a domain adaptation layer consists of a generator and a discriminator. The generator maps initial vegetation index features from the source domain to the target domain, ensuring a distribution similar to the target domain data. The discriminator distinguishes whether the features generated by the generator originate from the source or target domain. Through adversarial training between the generator and discriminator, the generator's parameters are continuously adjusted, allowing the generated features to better adapt to the target domain, thus achieving cross-domain feature transfer. The initial vegetation index features are input into a pre-trained feature transfer network. After high-dimensional feature space mapping processing by the domain adaptation layer, a transferred feature vector is generated. This transferred feature vector possesses better expressive power and transferability in the target domain, providing a crucial foundation for subsequent feature mapping and analysis.

[0040] Step S230: Obtain the preset dual-branch mapping model structure, process the transfer feature vector through the convolutional neural network architecture of the first branch to generate a convolutional feature map sequence, and perform temporal correlation modeling on the convolutional feature map sequence through the recurrent neural network architecture of the second branch to generate a temporal feature vector.

[0041] The pre-designed dual-branch mapping model structure consists of two branches, each handling different types of feature information. The first branch employs a convolutional neural network (CNN) architecture to automatically extract spatial features from the data. The second branch uses a recurrent neural network (RNN) architecture to capture temporal correlation information within the data.

[0042] The convolutional neural network (CNN) architecture processes the transfer feature vector as follows: First, the transfer feature vector is input into the initial convolutional layer of the CNN. The convolutional kernels in the convolutional layer perform convolution operations on the transfer feature vector, extracting local spatial features. Then, batch normalization and non-linear activation functions enhance the expressive power of the features. Next, through multiple convolutional and pooling layers, features are continuously extracted and compressed, generating a sequence of convolutional feature maps. This sequence of convolutional feature maps contains the spatial feature information from the transfer feature vector.

[0043] In one implementation, step S230 may specifically include the following steps S231-S235: Step S231: Reshape the transfer feature vector into a three-dimensional feature tensor, input it into the initial convolutional layer of the convolutional neural network architecture, and perform the first layer of feature extraction on the three-dimensional feature tensor through the convolutional kernel used to capture local spatial features to generate a primary convolutional feature map.

[0044] The transfer feature vector can be a one-dimensional vector, but to be processed in a convolutional neural network, it can be reshaped into a three-dimensional feature tensor. The three-dimensional feature tensor has three dimensions, representing height, width, and number of channels. During the reshaping process, the size of the three dimensions is appropriately allocated according to the length of the transfer feature vector and the input requirements of the convolutional neural network.

[0045] The initial convolutional layer in a convolutional neural network (CNN) architecture is the first layer and contains multiple convolutional kernels. A convolutional kernel is a small matrix that performs a sliding convolution operation on a 3D feature tensor to extract local spatial features. Each kernel performs convolution calculations on different regions of the 3D feature tensor, resulting in a feature map. Through the convolution operations of multiple kernels, multiple feature maps are generated, and these feature maps are combined to form the primary convolutional feature map.

[0046] Step S232: Perform batch normalization on the primary convolutional feature map, enhance the feature representation ability through a nonlinear activation function, and generate an activated primary convolutional feature map. The type of nonlinear activation function is determined according to the distribution characteristics of the vegetation index feature.

[0047] Batch normalization normalizes the input data to achieve a mean of 0 and a variance of 1. When performing batch normalization on the primary convolutional feature map, for each channel, the mean and variance of all pixels in that channel are calculated. Then, the mean is subtracted from the value of each pixel, and the result is divided by the standard deviation to obtain the normalized value. Non-linear activation functions introduce non-linearity to enhance the expressive power of the neural network. Possible non-linear activation functions include ReLU (Rectified Linear Unit), Sigmoid, and Tanh. The type of non-linear activation function is determined by the distribution characteristics of the vegetation index features. For example, if the vegetation index features are concentrated in the positive value region, ReLU can be chosen as the non-linear activation function. The expression for ReLU is: f(x) = max(0, x), which sets all negative values ​​to 0 and retains only positive values, effectively enhancing the expressive power of the features. After batch normalization of the primary convolutional feature map, the normalized feature map is input into the non-linear activation function for processing, resulting in the activated primary convolutional feature map. The activated primary convolutional feature map has a stronger feature representation capability and can better reflect the spatial feature information in the transferred feature vector.

[0048] Step S233: Input the activated primary convolutional feature map into the next convolutional layer, and use an expanded convolutional kernel with an expansion rate parameter to extract features with a larger receptive field to generate an intermediate convolutional feature map. The number of channels in the intermediate convolutional feature map is adjusted according to the feature expression requirements.

[0049] An dilated convolution kernel is a special type of convolution kernel that has a specific dilation parameter during the convolution operation. The dilation parameter determines the spacing between elements within the kernel. Compared to a regular convolution kernel, a dilated convolution kernel can expand the receptive field of the convolution operation without increasing the kernel size, thereby extracting a wider range of feature information.

[0050] When the activated primary convolutional feature map is input into the next convolutional layer, it is convolved using a dilated convolutional kernel with a dilation parameter. For example, when the dilation parameter is 2, the elements in the convolutional kernel are separated by one element during the convolution calculation. Through the convolution operation of the dilated convolutional kernel, feature information with a larger receptive field can be extracted from the activated primary convolutional feature map, generating an intermediate convolutional feature map.

[0051] The number of channels in an intermediate-level convolutional feature map is adjusted according to the feature representation requirements. More channels allow the intermediate-level convolutional feature map to express richer feature information, but also increase computational cost and model complexity. Therefore, the number of channels in the intermediate-level convolutional feature map needs to be determined reasonably based on specific feature representation requirements. For example, if more feature information needs to be extracted, the number of channels can be increased appropriately; if computational cost and model complexity are to be reduced, the number of channels can be decreased appropriately. Through feature extraction operations involving expanding the convolutional kernel, intermediate-level convolutional feature maps can contain a wider range of spatial feature information.

[0052] Step S234: Perform max pooling on the intermediate convolutional feature map to reduce the spatial dimension and retain key features, generating a dimension-reduced intermediate convolutional feature map. The pooling window size of the max pooling operation matches the spatial resolution of the intermediate convolutional feature map.

[0053] Max pooling reduces the spatial dimensionality of a feature map while preserving key feature information. When performing max pooling on an intermediate convolutional feature map, a pooling window size is first determined. Then, the pooling window slides across the intermediate convolutional feature map. For each pixel within the pooling window, the maximum value is taken as the output pixel value for that window. The pooling window size is matched to the spatial resolution of the intermediate convolutional feature map. If the spatial resolution of the intermediate convolutional feature map is high, a larger pooling window size can be chosen to reduce the spatial dimensionality more significantly; if the spatial resolution is low, a smaller pooling window size can be chosen to avoid excessive dimensionality reduction leading to the loss of key feature information. Through max pooling, the spatial dimensionality of the intermediate convolutional feature map is reduced while preserving its key feature information, generating a dimensionality-reduced intermediate convolutional feature map. The dimensionality-reduced intermediate convolutional feature map reduces computational cost and model complexity while still retaining sufficient feature information.

[0054] Step S235: Input the dimensionality-reduced intermediate convolutional feature map into the final convolutional layer, and perform feature optimization processing on it through a convolutional kernel that can achieve channel attention weighting to generate a sequence of convolutional feature maps containing multi-scale spatial features. The length of the convolutional feature map sequence is consistent with the number of sampling points in the time dimension of the transfer feature vector.

[0055] The final convolutional layer is the last layer in the convolutional neural network architecture. Its function is to perform feature optimization processing on the dimensionality-reduced intermediate convolutional feature maps, generating a sequence of convolutional feature maps containing multi-scale spatial features. A convolutional kernel capable of channel attention weighting is a special type of convolutional kernel that can weight features according to the importance of different channels, thereby highlighting key feature information.

[0056] When the reduced-dimensionality intermediate convolutional feature maps are input into the final convolutional layer, convolutional kernels that implement channel attention weighting are used to perform convolution operations on them. For example, the channel attention mechanism calculates the importance weight of each channel and then applies these weights to the convolutional kernel's convolution calculation, so that the feature information of important channels receives more attention and enhancement.

[0057] Through feature optimization processing in the final convolutional layer, a sequence of convolutional feature maps containing multi-scale spatial features is generated. Multi-scale spatial features refer to spatial feature information extracted at different scales, which can more comprehensively reflect the feature information of the data. The length of the convolutional feature map sequence is consistent with the number of sampling points in the time dimension of the transfer feature vector. This is to ensure that the convolutional feature map sequence can match the transfer feature vector in the time dimension. Through the processing of the final convolutional layer, the convolutional feature map sequence contains rich multi-scale spatial feature information.

[0058] Step S240: Input the temporal feature vector into the fusion output layer of the dual-branch mapping model structure, adjust the contribution weights of convolutional features and temporal features through the attention mechanism, generate a fusion feature vector, and determine the fusion feature vector as the cold resistance gene expression product feature to complete the feature transformation operation of the recessive stress resistance trait.

[0059] The fusion output layer of the dual-branch mapping model structure fuses convolutional features and temporal features to generate a comprehensive feature vector. The attention mechanism automatically adjusts the feature contribution weights, dynamically allocating the contribution weights of convolutional and temporal features based on their importance. When the temporal feature vector is input into the fusion output layer, the attention mechanism analyzes and evaluates the convolutional and temporal features, calculating the importance weight of each feature. Then, based on these weights, the convolutional and temporal features are weighted and summed to generate the fused feature vector.

[0060] For example, attention mechanisms can use attention score-based methods to calculate feature importance weights. First, a fully connected layer maps convolutional features and temporal features to a low-dimensional space, and then calculates their similarity scores. A higher similarity score indicates a more important feature, and its corresponding weight is larger. Finally, the convolutional and temporal features are weighted and summed based on the calculated weights to obtain a fused feature vector.

[0061] The fused feature vector was identified as the feature of the cold-resistance gene expression product, thus completing the feature transformation operation for recessive stress resistance traits. Recessive stress resistance traits refer to some trait characteristics of plants that are not easily observed directly during the stress resistance process. Through the processing of the bi-branch mapping model structure, vegetation index features were transformed into cold-resistance gene expression product features, thereby realizing the feature extraction and transformation of recessive stress resistance traits.

[0062] Step S300: Call the preset slope compensation model to perform terrain distortion correction processing on the terrain structure data in the spatiotemporal correlation dataset to obtain standardized terrain data that eliminates the influence of slope. Combine the standardized terrain data to perform spatial feature calibration processing on the characteristics of cold resistance gene expression products.

[0063] The preset slope compensation model is a pre-designed model used to correct for terrain distortion in terrain structure data. Terrain distortion refers to the errors and deformations that occur in terrain structure data during acquisition and processing due to the presence of terrain slope.

[0064] Terrain distortion correction is a crucial step in processing terrain structure data within a spatiotemporally correlated dataset. Its purpose is to eliminate the influence of terrain slope on the terrain structure data, resulting in standardized terrain data free from slope effects. During terrain distortion correction, the terrain structure data is input into a pre-defined slope compensation model. The model then performs correction calculations based on the terrain slope information, thereby eliminating the impact of terrain distortion.

[0065] Spatial feature calibration is a process of processing the characteristics of cold-resistance gene expression products by combining standardized topographic data. Standardized topographic data contains accurate topographic information. By spatially registering and associating the characteristics of cold-resistance gene expression products with standardized topographic data, the characteristics of cold-resistance gene expression products can be calibrated, making them more accurately reflect the stress resistance characteristics of plants under different topographic conditions.

[0066] For example, in high-altitude, steep slope regions, different terrain gradients can affect the growth environment and stress resistance of plants. By performing terrain distortion correction on the terrain structure data to obtain standardized terrain data, and then combining the standardized terrain data with spatial feature calibration of the cold resistance gene expression product characteristics, we can more accurately analyze the stress resistance performance of plants under different terrain conditions, providing a more scientific basis for plant selection and cultivation.

[0067] In one implementation, step S300 may specifically include the following steps S310-S350: Step S310: Extract elevation and slope data from the topographic structure data in the spatiotemporal correlation dataset, construct a digital elevation model, and extract the slope value distribution matrix from the digital elevation model using a slope calculation algorithm.

[0068] Elevation data describes the height of a terrain structure, while slope data describes the degree of inclination of the terrain. A digital elevation model (DEM) is a model that represents the elevation of a terrain surface in digital form, and it can be obtained by interpolating and processing elevation data.

[0069] After extracting elevation and slope data from the topographic structure data in the spatiotemporally correlated dataset, interpolation algorithms are used to convert the discrete elevation data into a continuous elevation surface, constructing a digital elevation model (DEM). Possible interpolation algorithms include bilinear interpolation and Kriging interpolation. Slope calculation algorithms are used to extract slope values ​​from the DEM. For example, a slope calculation algorithm based on the finite difference method calculates the elevation difference between each pixel in the DEM and its neighboring pixels, and then calculates the slope value of that pixel based on the elevation difference and the distance between the pixels.

[0070] The digital elevation model is processed by a slope calculation algorithm to obtain the slope value of each pixel. These slope values ​​are then combined into a matrix, namely the slope value distribution matrix. The slope value distribution matrix can intuitively reflect the slope distribution of the terrain, providing important basic data for subsequent terrain distortion correction processing.

[0071] Step S320: Input the slope value distribution matrix into the terrain factor calculation layer of the slope compensation model, calculate the slope change rate and aspect variation coefficient through the terrain curvature analysis algorithm, and generate the terrain distortion influence factor.

[0072] The topographic factor calculation layer in the slope compensation model calculates the topographic distortion influence factor based on the slope value distribution matrix. The topographic curvature analysis algorithm is used to analyze topographic curvature, calculating the slope change rate and aspect coefficient of variation. The slope change rate refers to the rate of change of the topographic slope in space, reflecting the degree of topographic undulation. The aspect coefficient of variation refers to the degree of change of the topographic aspect in space, reflecting the directional change of the topographic slope.

[0073] In one implementation, step S320 may specifically include the following steps S321-S325: Step S321: Perform a first-order difference operation on the slope value distribution matrix to calculate the slope gradient values ​​in the horizontal and vertical directions, and construct the slope gradient matrix.

[0074] First-order differencing is used to calculate the rate of change of a function between adjacent points. When performing first-order differencing on the slope value distribution matrix, the differences in the horizontal and vertical directions are calculated separately. For the horizontal difference, for each element in the slope value distribution matrix, the difference between it and its right-hand neighbor is calculated to obtain the horizontal slope gradient value. For the vertical difference, for each element in the slope value distribution matrix, the difference between it and its lower-hand neighbor is calculated to obtain the vertical slope gradient value. The horizontal and vertical slope gradient values ​​are combined to construct the slope gradient matrix. Each element in the slope gradient matrix represents the rate of change of the slope at that location in both the horizontal and vertical directions.

[0075] For example, consider a two-dimensional slope distribution matrix, assuming its size is M x N. When performing a horizontal difference operation, for the element in the i-th row and j-th column of the matrix, its horizontal slope gradient is the difference between that element and the element in the i-th row and (j+1)-th column. Similarly, when performing a vertical difference operation, for the element in the i-th row and j-th column of the matrix, its vertical slope gradient is the difference between that element and the element in the (i+1)-th row and j-th column.

[0076] Step S322: Perform second-order differential processing on the slope gradient matrix using the terrain curvature analysis algorithm to calculate the profile curvature and planar curvature.

[0077] The terrain curvature analysis algorithm is used to analyze terrain curvature. It calculates the profile curvature and planar curvature of the terrain by performing a second-order differential on the slope gradient matrix. Profile curvature refers to the curvature of the terrain in the vertical direction, reflecting its vertical undulations. Planar curvature refers to the curvature of the terrain in the horizontal direction, reflecting its horizontal bending. When performing a second-order differential on the slope gradient matrix, the second-order differences of the slope gradient matrix in the horizontal and vertical directions are calculated separately. For the second-order difference in the horizontal direction, the difference of the horizontal slope gradient values ​​in the slope gradient matrix is ​​calculated in the horizontal direction. For the second-order difference in the vertical direction, the difference of the vertical slope gradient values ​​in the slope gradient matrix is ​​calculated in the vertical direction. Based on the results of the second-order differences in the horizontal and vertical directions, the profile curvature and planar curvature are calculated. The specific calculation method can be performed according to the specific formulas of the terrain curvature analysis algorithm.

[0078] Step S323: Calculate the slope change rate based on the profile curvature and the plane curvature. The slope change rate is the weighted sum of the absolute values ​​of the profile curvature and the plane curvature.

[0079] The slope change rate is an indicator reflecting the rate of change of terrain slope in space, comprehensively considering the curvature changes of the terrain in both the vertical and horizontal directions. When calculating the slope change rate based on profile curvature and planar curvature, the absolute values ​​of profile curvature and planar curvature are first taken, and then they are summed according to preset weighting coefficients. For example, assuming the weighting coefficient for profile curvature is w1, the weighting coefficient for planar curvature is w2, and w1 + w2 = 1, then the formula for calculating the slope change rate is: Slope Change Rate = w1 × |Profile Curvature| + w2 × |Planar Curvature|. The values ​​of the weighting coefficients can be adjusted according to specific terrain features and research needs. If more attention is paid to the vertical changes in terrain, the weighting coefficient for profile curvature can be appropriately increased; if more attention is paid to the horizontal changes in terrain, the weighting coefficient for planar curvature can be appropriately increased.

[0080] Step S324: Extract the slope aspect from the terrain structure data in the spatiotemporal correlation dataset to generate a slope aspect angle matrix. Calculate the local standard deviation of the slope aspect angle matrix using a sliding window algorithm and determine the local standard deviation as the slope aspect variation coefficient.

[0081] Slope aspect refers to the direction of inclination of a terrain surface. Slope aspect extraction is the process of extracting slope aspect information from terrain structure data. The slope aspect angle can be determined by calculating the normal vector of the terrain surface. For each pixel in the terrain structure data, the direction of its normal vector is calculated, and this direction is converted into an angle value to obtain the slope aspect angle of that pixel. The slope aspect angles of all pixels are then combined into a matrix, namely the slope aspect angle matrix.

[0082] The sliding window algorithm is used to calculate local statistics. When calculating the local standard deviation of the aspect angle matrix, a sliding window size is first determined. Then, the sliding window is moved across the aspect angle matrix. For each aspect angle value within the sliding window, its standard deviation is calculated. The standard deviation of each sliding window is used as the local standard deviation of the central pixel of that window. The local standard deviations of all pixels are combined into a matrix, namely the aspect coefficient of variation matrix. The aspect coefficient of variation reflects the degree of spatial variation of the aspect, providing important parameters for subsequent calculations of topographic distortion influence factors.

[0083] By extracting and processing the slope aspect and calculating using the sliding window algorithm, the slope aspect variation coefficient is obtained, which can more accurately reflect the slope aspect changes of the terrain.

[0084] Step S325: Input the slope change rate and aspect variation coefficient into the preset terrain distortion influence factor algorithm, and merge the two parameters into a terrain distortion influence factor through normalization processing.

[0085] The preset terrain distortion impact factor algorithm is an algorithm that integrates the slope change rate and the aspect variation coefficient. The purpose is to generate a comprehensive terrain distortion impact factor to describe the degree of distortion impact of terrain slope and aspect on terrain structure data.

[0086] When inputting the slope change rate and aspect variability coefficient into the preset terrain distortion impact factor algorithm, these two parameters are first normalized, such as using the min-max normalization method and the Z-score normalization method. After normalizing the slope change rate and aspect variability coefficient, the normalized slope change rate and aspect variability coefficient are weighted and summed according to the preset formula of the terrain distortion impact factor algorithm to obtain the terrain distortion impact factor. The terrain distortion impact factor can comprehensively reflect the degree of distortion influence of terrain slope and aspect on terrain structure data.

[0087] Step S330: Construct a spatial correction function based on the terrain distortion influence factor, perform geometric distortion correction on the digital elevation model, and generate standardized terrain data that eliminates the influence of slope.

[0088] A spatial correction function is a function constructed based on the terrain distortion influencing factor. Its purpose is to correct geometric distortions in digital elevation models, eliminating the influence of terrain slope on terrain data. When constructing a spatial correction function based on the terrain distortion influencing factor, the function type and parameters are first determined. One type of spatial correction function is a linear function, with the form: f(x) = a × x + b, where x is the terrain distortion influencing factor, and a and b are the function parameters. Parameters a and b can be determined experimentally or through fitting methods, ensuring that the spatial correction function can effectively correct terrain distortion.

[0089] In one implementation, step S330 may specifically include the following steps S331-S335: Step S331: Construct a spatial correction function with the topographic distortion influencing factor as the independent variable. The function value of the spatial correction function is positively correlated with the topographic distortion influencing factor.

[0090] The spatial correction function is used to correct geometric distortions in digital elevation models. It uses the terrain distortion factor as its independent variable. The terrain distortion factor reflects the degree of distortion caused by terrain slope and aspect on terrain structure data. The function value of the spatial correction function is positively correlated with the terrain distortion factor, meaning that the larger the terrain distortion factor, the larger the function value of the spatial correction function, and the greater the degree of correction to the terrain data.

[0091] As mentioned earlier, the spatial correction function can be a linear function, expressed as: f(x) = a × x + b, where x is the terrain distortion factor, a and b are the parameters of the function, and a > 0 to ensure a positive correlation between the function value and the terrain distortion factor. The parameters a and b can be determined experimentally or through fitting methods. For example, a set of known terrain distortion factors and corresponding corrected terrain data can be selected, and the values ​​of parameters a and b can be fitted using the least squares method, so that the spatial correction function can correct terrain distortion as accurately as possible.

[0092] Besides linear functions, other types of functions can be chosen as spatial correction functions, such as quadratic functions and exponential functions. Different types of functions have different characteristics and applicable ranges, and can be selected according to specific terrain features and correction requirements.

[0093] Step S332: Perform correction calculations on each elevation sampling point in the digital elevation model. Multiply the original elevation value of the corresponding elevation sampling point by the function value of the spatial correction function at that sampling point location to obtain the preliminary corrected elevation value.

[0094] After constructing the spatial correction function, correction calculations are performed on each elevation sampling point in the digital elevation model. The digital elevation model consists of a series of elevation sampling points, each with a corresponding original elevation value.

[0095] For each elevation sampling point in the digital elevation model, the terrain distortion influence factor for that sampling point is first determined. Then, this terrain distortion influence factor is substituted into the spatial correction function to calculate the function value of the spatial correction function at that sampling point.

[0096] Finally, the original elevation value of the sampling point is multiplied by the function value of the spatial correction function at that sampling point location to obtain the preliminary corrected elevation value. For example, if the original elevation value of a certain elevation sampling point is h, the terrain distortion influence factor of the sampling point is x, and the spatial correction function is f(x), then the preliminary corrected elevation value is h' = h × f(x).

[0097] By performing correction calculations on each elevation sampling point, preliminary corrected elevation values ​​are obtained. These preliminary corrected elevation values ​​can eliminate the influence of terrain slope on terrain data to a certain extent, but there may still be some influence of slope inclination on elevation measurement values, which require further secondary correction processing.

[0098] Step S333: Perform secondary correction on the preliminary corrected elevation value based on the slope value of the sampling point. Use the slope cosine correction algorithm to eliminate the influence of slope inclination on the elevation measurement value and generate the secondary corrected elevation value.

[0099] Slope inclination can introduce errors into elevation measurements. To eliminate these errors, a secondary correction is needed on the initially corrected elevation values. The slope cosine correction algorithm is a commonly used secondary correction algorithm. It calculates the slope cosine value based on the slope value of the sampling points, and then divides the initially corrected elevation value by the slope cosine value to obtain the secondary corrected elevation value.

[0100] When performing the slope cosine correction algorithm, the slope values ​​of the sampling points are first obtained. These slope values ​​can be obtained from the previously constructed slope value distribution matrix. Then, the slope cosine value is calculated based on the slope values. Finally, the initially corrected elevation value is divided by the slope cosine value to obtain the second-corrected elevation value. This second-correction process effectively eliminates the influence of slope inclination on the elevation measurement, making the second-corrected elevation value more accurately reflect the true height of the terrain.

[0101] Step S334: Reconstruct the digital elevation model by reconstructing the elevation values ​​of all elevation sampling points after secondary correction, and generate standardized terrain data that eliminates the influence of slope.

[0102] After performing a secondary correction on the initial corrected elevation values ​​of all elevation sampling points, the secondary corrected elevation values ​​are obtained. These secondary corrected elevation values ​​are then used to reconstruct the digital elevation model, generating standardized terrain data that eliminates the influence of slope.

[0103] The process of reconstructing a digital elevation model (DEM) is similar to the previous process of constructing a DEM. Interpolation algorithms can be used to convert discrete, quadratically corrected elevation values ​​into a continuous elevation surface. Common interpolation algorithms include bilinear interpolation and kriging interpolation.

[0104] By reconstructing the digital elevation model, the standardized terrain data eliminates the influence of terrain slope on the terrain data, and can more accurately reflect the true terrain situation. The standardized terrain data provides reliable basic data for subsequent spatial feature calibration processing, enabling the characteristics of cold-resistant gene expression products to be more accurately correlated and analyzed with terrain information.

[0105] Step S335: Adjust the spatial resolution of the standardized terrain data to match the spatial resolution of the cold-resistance gene expression product characteristics.

[0106] Spatial resolution refers to the precision and detail of data in space. Standardized terrain data and cold-resistance gene expression product characteristics may have different spatial resolutions. In order to effectively correlate and analyze them, the spatial resolution of the standardized terrain data needs to be adjusted to match that of the cold-resistance gene expression product characteristics.

[0107] When adjusting the spatial resolution of standardized terrain data, interpolation or downsampling methods can be used. If the spatial resolution of the standardized terrain data is lower than that of the characteristics of the cold-resistance gene expression product, interpolation algorithms can be used to improve its spatial resolution. Feasible interpolation algorithms include bilinear interpolation and cubic spline interpolation. If the spatial resolution of the standardized terrain data is higher than that of the characteristics of the cold-resistance gene expression product, downsampling algorithms can be used to reduce its spatial resolution. Common downsampling algorithms include average pooling and max pooling.

[0108] By adjusting the spatial resolution, standardized topographic data and cold-resistant gene expression product characteristics can have the same spatial resolution, enabling more accurate analysis of plant resistance under different topographic conditions.

[0109] Step S340: Extract the spatial coordinate information of the cold resistance gene expression product characteristics, perform spatial registration processing on the spatial coordinate information and standardized terrain data, and establish the spatial correspondence between the cold resistance gene expression product characteristics and the standardized terrain data.

[0110] Spatial coordinate information of cold-resistance gene expression product characteristics refers to the location information of these characteristics in geographic space, which can be obtained from the feature's metadata. Spatial registration is the process of matching and aligning the spatial coordinate information of cold-resistance gene expression product characteristics with standardized terrain data. Its purpose is to establish a spatial correspondence between cold-resistance gene expression product characteristics and standardized terrain data, so that the two are spatially consistent.

[0111] In one implementation, step S340 may specifically include the following steps S341 - S345: Step S341: Parse the original acquisition coordinate system from the metadata of the cold-resistant gene expression product characteristics, and convert the spatial coordinate information into the same geographic coordinate system as the standardized terrain data through a coordinate transformation algorithm. The parameters of the coordinate transformation algorithm are determined according to the projection method of the two coordinate systems.

[0112] The metadata of cold-resistance gene expression product characteristics includes the original acquisition coordinate system information for that characteristic. The original acquisition coordinate system may be different projection methods, such as UTM projection, Gauss-Kruger projection, etc., while standardized terrain data usually adopts a unified geographic coordinate system, such as WGS84 projection.

[0113] When retrieving the original acquisition coordinate system from the metadata of cold-resistance gene expression product characteristics, it is necessary to parse and extract the metadata. Metadata is usually stored in a specific format, such as XML or JSON. Appropriate parsing tools can be used to parse the metadata and extract the information of the original acquisition coordinate system.

[0114] Coordinate transformation algorithms are used to convert spatial coordinate information from one coordinate system to another. The parameters of a coordinate transformation algorithm are determined by the projection method of the two coordinate systems. For example, if the original acquisition coordinate system is UTM projection and the geographic coordinate system of the standardized terrain data is WGS84 projection, then a UTM to WGS84 projection transformation algorithm is needed. The parameters of this algorithm include information such as the UTM projection zone number and the central meridian.

[0115] During coordinate transformation, the spatial coordinate information of the cold-resistance gene expression product characteristics is substituted into the coordinate transformation algorithm. Following the algorithm's calculation steps, this information is converted into a geographic coordinate system identical to the standardized terrain data. This coordinate transformation ensures that the spatial coordinate information of the cold-resistance gene expression product characteristics is consistent with the standardized terrain data in the geographic coordinate system.

[0116] Step S342: Calculate the Euclidean distance between the transformed spatial coordinate information and the center coordinates of the standardized terrain data grid cells, and determine the initial grid cell to which each feature point belongs based on the principle of minimum distance.

[0117] Standardized terrain data is typically divided into multiple grid cells, each with a central coordinate. When calculating the Euclidean distance between the transformed spatial coordinates and the central coordinates of the standardized terrain data grid cells, for each feature point, the transformed spatial coordinates are used to calculate its Euclidean distance to the central coordinates of all grid cells. Based on the principle of minimum distance, for each feature point, the central coordinates of the grid cell with the smallest Euclidean distance are found, and the feature point is assigned to that grid cell as its initial grid cell. For example, if a feature point has the smallest Euclidean distance to grid cell A, then the feature point is assigned to grid cell A. By calculating the Euclidean distance and determining the initial grid cell based on the principle of minimum distance, a preliminary correspondence between the characteristics of cold-resistance gene expression products and the standardized terrain data grid cells can be established.

[0118] Step S343: Fine-tune the coordinates of feature points at the boundary of the initial grid cell using a bidirectional linear interpolation algorithm, correct the systematic errors in the coordinate transformation process, and generate fine-tuned spatial coordinates. The interpolation weights of the bidirectional linear interpolation algorithm are calculated based on the distance from the feature point to the grid boundary.

[0119] Systematic errors may exist during coordinate transformation, resulting in inaccurate correspondence between feature point coordinates and standardized terrain data grid cells. To correct these systematic errors, the coordinates of feature points at the boundaries of the initial grid cells need to be fine-tuned.

[0120] Bidirectional linear interpolation is a coordinate fine-tuning algorithm that calculates interpolation weights based on the distance from a feature point to the grid boundary, thereby adjusting the coordinates of the feature point. In performing bidirectional linear interpolation, the grid cell containing the feature point and its adjacent grid cells are first determined. Then, based on the distance from the feature point to the grid boundary, the interpolation weight of that feature point in the adjacent grid cells is calculated. Through the coordinate fine-tuning process of bidirectional linear interpolation, systematic errors in the coordinate transformation process can be corrected, making the fine-tuned spatial coordinates more accurately correspond to the grid cells of the standardized terrain data.

[0121] Step S344: Construct a spatial registration error evaluation function to calculate the root mean square error between the fine-tuned spatial coordinates and the standardized terrain data grid coordinates. When the error value exceeds the preset error threshold, perform coordinate transformation and fine-tuning operations again until the error meets the requirements.

[0122] The spatial registration error evaluation function is used to assess the accuracy of spatial registration. It measures the degree of error by calculating the root mean square error (RMSE) between the fine-tuned spatial coordinates and the coordinates of the standardized terrain data grid. When constructing the spatial registration error evaluation function, the fine-tuned spatial coordinates and the standardized terrain data grid coordinates are substituted into the RMSE calculation formula to calculate the error value. The preset error threshold is an upper limit set based on actual needs and experience. When the calculated error value exceeds the preset error threshold, it indicates that the accuracy of spatial registration is insufficient, and coordinate transformation and fine-tuning operations need to be repeated. When re-performing coordinate transformation, the parameters of the coordinate transformation algorithm can be adjusted, or other more accurate coordinate transformation algorithms can be used. When re-performing fine-tuning, the parameters of the bidirectional linear interpolation algorithm can be adjusted, or other more suitable fine-tuning algorithms can be used. By continuously repeating coordinate transformation and fine-tuning operations until the calculated error value meets the requirements of the preset error threshold, the accuracy of spatial registration has reached the expected standard.

[0123] Step S345: Based on the correspondence between the finely adjusted spatial coordinates and grid cells, establish a spatial index table of cold-resistant gene expression product characteristics and standardized terrain data. The spatial index table is used to realize the rapid query and association of feature points and terrain grid cells.

[0124] A spatial index table is a data structure used for quickly querying and associating feature points with terrain grid cells. When building a spatial index table based on the finely adjusted correspondence between spatial coordinates and grid cells, the structure and storage method of the index table must first be determined.

[0125] One spatial index table structure is, for example, a hash table, which uses the identifier of a grid cell as the key and the list of feature points contained in that grid cell as the value. For each refined spatial coordinate, based on its grid cell, the relevant information of the feature point (such as feature value, coordinates, etc.) is added to the corresponding grid cell list.

[0126] For example, iterate through all the finely tuned spatial coordinates, and for each coordinate, determine its corresponding grid cell identifier. Then, look up the list corresponding to that grid cell identifier in the hash table. If the list does not exist, create a new list and add the information of that feature point to the list; if the list already exists, add the information of that feature point to the end of the list.

[0127] In this way, a spatial index table of cold-resistance gene expression product characteristics and standardized terrain data was established. The establishment of the spatial index table enables the use of the hash table's fast lookup function to obtain results in a short time when it is necessary to query the terrain grid cell to which a certain feature point belongs, or to query the feature points contained in a certain terrain grid cell, greatly improving the efficiency of querying and associating feature points with terrain grid cells.

[0128] Step S350: Spatial interpolation is performed on the characteristics of cold-resistant gene expression products according to the spatial correspondence. The inverse distance weighted interpolation algorithm is used to map the characteristic values ​​of cold-resistant gene expression products to the grid cells of standardized terrain data.

[0129] The spatial correspondence is established through the preceding spatial registration process, which clarifies the correspondence between the characteristics of cold-resistance gene expression products and the standardized terrain data grid cells. Spatial interpolation transforms the discrete cold-resistance gene expression product characteristic values ​​into a continuous spatial distribution, in order to gain a more comprehensive understanding of the distribution of these characteristics across the terrain.

[0130] Inverse distance weighted interpolation is a spatial interpolation algorithm that determines the value of an interpolation point based on the inverse relationship between distances. The basic idea of ​​the algorithm is that the value of an interpolation point is the weighted average of the values ​​of surrounding known points, with the weights inversely proportional to the distances from the known points to the interpolation point.

[0131] When mapping the feature values ​​of cold-resistance gene expression products to grid cells of standardized terrain data using an inverse distance weighted interpolation algorithm, for each grid cell center (i.e., the interpolation point) in the standardized terrain data, the known feature points surrounding it are first determined. Then, the distance from each known feature point to the interpolation point is calculated. Next, the weight of each known feature point is calculated based on the inverse relationship of distance, with points that are closer to each other having a larger weight. Finally, the feature value of each known feature point is multiplied by its corresponding weight, and all products are summed to obtain the value of the interpolation point.

[0132] By using spatial interpolation with inverse distance weighted interpolation algorithm, the feature values ​​of cold-resistant gene expression products are mapped to grid cells of standardized terrain data, thus obtaining the continuous distribution of cold-resistant gene expression product features on the terrain.

[0133] Step S400: Perform time-series trend analysis based on the characteristics of cold-resistant gene expression products after spatial feature calibration and environmental monitoring data in the spatiotemporal correlation dataset. Calculate growth adaptability indicators for different time periods using a plant growth adaptability assessment algorithm, and generate plant growth adaptability labels based on these indicators.

[0134] Temporal trend analysis is a method for studying the patterns of data changes over time. By performing temporal trend analysis on the characteristics of cold-resistant gene expression products after spatial feature calibration and environmental monitoring data in spatiotemporally correlated datasets, we can understand the changes in plant stress resistance traits and environmental factors over time.

[0135] Plant growth adaptability assessment algorithms are used to evaluate the ability of plants to adapt to different environmental conditions. Combining the characteristics of cold-resistance gene expression products and environmental monitoring data, they calculate growth adaptability indices at different time periods. A growth adaptability index is a comprehensive numerical value reflecting the degree of plant adaptation to the environment. Generating plant growth adaptability labels based on these indices involves classifying and labeling them to provide a more intuitive understanding of plant growth adaptation. For example, growth adaptability indices can be divided into different levels, such as "well adapted," "moderately adapted," and "poorly adapted," and each level can be assigned a corresponding label. Through time-series trend analysis, growth adaptability index calculation, and growth adaptability label generation, a comprehensive assessment of plant growth adaptability in high-altitude, cold, and steep slope environments can be achieved.

[0136] In one implementation, step S400 may specifically include the following steps S410-S450: Step S410: Extract environmental monitoring data from the spatiotemporal correlation dataset. Each type of environmental monitoring data has timestamp information.

[0137] Environmental monitoring data is crucial for reflecting the environmental conditions of high-altitude, steep slope regions, including various types such as temperature, humidity, light intensity, and soil moisture. Timestamps record the data collection time, giving environmental monitoring data a temporal dimension and facilitating time-series analysis.

[0138] When extracting environmental monitoring data from a spatiotemporally correlated dataset, the dataset first needs to be filtered and categorized. Spatiotemporally correlated datasets contain various types of data, including topographic structure data, vegetation image data, and environmental monitoring data. Environmental monitoring data needs to be filtered based on the data type and characteristics. Simultaneously, to ensure that all types of environmental monitoring data contain timestamp information, the accurate collection time for each data point should be recorded during the data acquisition process. When extracting environmental monitoring data, it is necessary to check whether the data contains timestamp information; if timestamp information is missing, it needs to be supplemented or corrected.

[0139] In one implementation, step S410 may specifically include the following steps S411-S415: Step S411: Perform data integrity checks on the environmental monitoring data in the spatiotemporal correlation dataset, identify missing data points and abnormal data points, use a linear interpolation algorithm to fill in the missing data points, and use the 3σ criterion to replace the abnormal data points to generate a complete environmental monitoring dataset.

[0140] Data integrity testing is a crucial step in ensuring the quality of environmental monitoring data. It identifies missing and outlier data points. Missing data points refer to data records lacking certain time points or data types, while outlier data points are those whose values ​​significantly deviate from the normal range.

[0141] Linear interpolation algorithms estimate the value of missing data points based on the values ​​of neighboring data points, using a linear relationship. For a missing data point, two neighboring known data points are found, and the value of the missing data point is calculated using a linear interpolation formula based on the values ​​of these two known data points and the time interval between them and the missing data point.

[0142] The 3σ criterion is based on the mean and standard deviation of data. Data values ​​within the range of the mean plus or minus three standard deviations are considered normal, while those outside this range are considered outliers. For identified outlier data points, they can be replaced with the average of the adjacent normal data points or other reasonable values.

[0143] By performing data integrity checks, missing data imputation, and abnormal data replacement, a complete environmental monitoring dataset is generated, thereby improving the quality and usability of environmental monitoring data.

[0144] Step S412: Classify and extract the complete environmental monitoring dataset according to the data collection type, and extract different types of environmental datasets. Each dataset contains a timestamp field and a monitoring value field.

[0145] Different types of environmental monitoring data have different physical meanings and characteristics. To facilitate subsequent analysis and processing, it is necessary to classify and extract complete environmental monitoring datasets according to the data collection type. For example, environmental monitoring data can be divided into temperature datasets, humidity datasets, light intensity datasets, soil moisture datasets, etc.

[0146] During the classification and extraction process, it is essential to ensure that each extracted dataset includes a timestamp field and a monitoring value field. The timestamp field records the time of data collection, while the monitoring value field records the specific monitoring data value. This way, each dataset possesses both time and data value dimensions, facilitating time-series analysis and data mining.

[0147] Step S413: Perform time format unification processing on various types of environmental datasets, converting all timestamp information into a unified time encoding format.

[0148] Different environmental monitoring devices or systems may use different time formats to record timestamp information, which can cause difficulties for subsequent data analysis and processing. Therefore, it is necessary to unify the time format of various types of environmental datasets and convert all timestamp information into a unified time encoding format.

[0149] Feasible time encoding formats include ISO 8601 and Unix timestamps. When unifying time formats, a suitable encoding format can be selected based on actual needs. For timestamp information in different formats, time processing functions provided by programming languages ​​or tools can be used for conversion. Through unified time format processing, the timestamp information of various types of environmental datasets becomes consistent.

[0150] Step S414: Perform time series resampling processing on each type of environmental dataset. Perform interpolation or downsampling operations on the environmental monitoring data according to the preset time granularity to keep the time sampling interval of each type of environmental data consistent.

[0151] Time series resampling is an operation that adjusts the time interval of time series data. It can convert environmental monitoring data with different sampling intervals into data with the same time sampling interval, facilitating data comparison and analysis. The preset time granularity is a time interval set according to actual needs and analysis objectives, such as hourly or daily. When performing time series resampling, if the time sampling interval of an environmental dataset is smaller than the preset time granularity, downsampling is required, i.e., selecting a subset of data points to represent data within a certain period. If the time sampling interval of an environmental dataset is larger than the preset time granularity, interpolation is required, i.e., estimating the data values ​​of missing time points based on known data point values.

[0152] Interpolation can use linear interpolation algorithms or other more complex interpolation algorithms, while downsampling can choose to take the average, maximum, or minimum values ​​to represent data over a period of time. Time series resampling ensures that the time sampling intervals for different types of environmental data remain consistent.

[0153] Step S415: Perform data standardization processing on the resampled environmental monitoring data to convert the monitoring values ​​to a preset data range.

[0154] Data standardization transforms monitoring values ​​from different ranges and scales into a unified data range to facilitate data comparison and analysis. The preset data range can be set according to actual needs and data characteristics, such as [0, 1] or [-1, 1]. Through data standardization, resampled environmental monitoring data is transformed into a preset data range, eliminating the influence of data scale and improving data comparability and model stability.

[0155] Step S420: Perform time dimension sampling processing on the cold resistance gene expression product characteristics after spatial feature calibration, extract the cold resistance gene expression product characteristic values ​​at different collection times according to the preset time interval, and generate a time-series-based stress resistance trait characteristic sequence.

[0156] The spatial feature calibration process for cold-resistance gene expression products incorporates information from both spatial and temporal dimensions. Temporal sampling involves extracting feature values ​​from these features at preset time intervals, converting them into a time-series format.

[0157] The preset time interval should be consistent with the time granularity of the environmental monitoring data during time series resampling to ensure subsequent time axis alignment. During time-dimensional sampling, the cold-resistance gene expression product features after spatial feature calibration are traversed, and feature values ​​are extracted according to the preset time interval based on the timestamp information.

[0158] For example, if the preset time interval is one hour, then the characteristic values ​​of cold-resistance gene expression products for each hour are extracted from the feature data, and these characteristic values ​​are arranged in chronological order to generate a time-series-based sequence of stress resistance trait characteristics. This sequence can clearly reflect the changes in cold-resistance gene expression product characteristics over time, providing an important data foundation for subsequent joint analysis.

[0159] Step S430: Align the environmental monitoring data with the stress resistance trait sequence on the time axis to generate a spatiotemporally aligned joint analysis dataset.

[0160] Time axis alignment matches environmental monitoring data and stress resistance trait characteristic sequences along the time dimension, so that they have corresponding data values ​​at the same point in time, facilitating joint analysis.

[0161] As one implementation method, step S430 involves aligning the environmental monitoring data with the stress resistance trait characteristic sequence along the time axis to generate a spatiotemporally aligned joint analysis dataset, including S431-S435: Step S431: Extract the timestamp information of each data point from the environmental monitoring data to construct an environmental timestamp sequence. At the same time, extract the corresponding timestamp information from the stress resistance trait characteristic sequence to construct a trait timestamp sequence.

[0162] The environmental timestamp sequence is a sequence composed of timestamp information of each data point in the environmental monitoring data, while the trait timestamp sequence is a sequence composed of timestamp information of each data point in the stress resistance trait characteristic sequence. By extracting these two sequences separately, the temporal distribution of the environmental monitoring data and the stress resistance trait characteristic sequence can be clearly understood.

[0163] Step S432: Perform time reference unification processing on the environmental timestamp sequence and the phenotype timestamp sequence, converting all timestamps into a time encoding format with the same precision to ensure a consistent time reference system.

[0164] Environmental timestamp sequences and phenotype timestamp sequences may use different time encoding formats or have different time precisions, which can affect the accuracy of timeline alignment. Therefore, it is necessary to unify the time base of these two sequences, converting all timestamps to a time encoding format with the same precision. A unified time encoding format, such as ISO8601 or Unix timestamps, can be chosen, and the corresponding time processing functions can be used to convert the timestamps in the environmental and phenotype timestamp sequences to this format. At the same time, it is essential to ensure consistent time precision, for example, accurate to seconds or milliseconds.

[0165] Step S433: Use set intersection operation to obtain the common timestamp set of the environmental timestamp sequence and the phenotype timestamp sequence, and use the common timestamp set as the reference time point for time axis alignment.

[0166] Set intersection is an operation used to find common elements in two sets. In timeline alignment, set intersection is used to obtain the common timestamp set of the environmental timestamp sequence and the trait timestamp sequence. These common timestamps are the time points that exist in both sequences. Using the common timestamp set as the reference time point for timeline alignment means that only the data corresponding to the common timestamps in the environmental monitoring data and the stress resistance trait sequence are retained, while data from other time points are ignored. This ensures that the environmental monitoring data and the stress resistance trait sequence have corresponding data values ​​at the same time points, facilitating joint analysis.

[0167] Step S434: Based on the common timestamp set, filter the environmental monitoring data and stress resistance trait sequence, retain the environmental data and trait values ​​corresponding to the common timestamp, and use a linear interpolation algorithm to supplement the missing environmental data and trait values ​​in the common timestamp set.

[0168] After obtaining the common timestamp set, it is necessary to filter the environmental monitoring data and stress resistance trait characteristic sequences based on this set. For the environmental monitoring data and stress resistance trait characteristic sequences, only the environmental data and trait value corresponding to the common timestamp are retained.

[0169] In practice, there may be instances where environmental data or trait values ​​are missing from certain time points in a common timestamp set. Linear interpolation algorithms are needed to supplement these missing values. The principle of linear interpolation is the same as the method used to handle missing values ​​in environmental monitoring data: it estimates the missing values ​​based on the values ​​of adjacent known data points and the time interval.

[0170] By filtering data and filling in missing values, we ensured that both environmental monitoring data and stress resistance trait sequences had complete data values ​​at the time points corresponding to the common timestamps, thus providing a guarantee for generating spatiotemporally aligned joint analysis datasets.

[0171] Step S435: Combine the filtered and supplemented environmental monitoring data with the stress resistance trait sequence according to the common timestamp order to generate a spatiotemporally aligned joint analysis dataset. Each record in the joint analysis dataset contains environmental data and trait value corresponding to the timestamp.

[0172] After data filtering and missing value imputation, the filtered and imputed environmental monitoring data are combined with the stress resistance trait sequence according to common timestamp order. Environmental data and trait values ​​corresponding to the same common timestamp are grouped together to form a single record. Thus, each record in the generated spatiotemporally aligned joint analysis dataset contains environmental data and trait values ​​at the corresponding timestamp, providing information in both the temporal and data dimensions. This dataset can be used for subsequent multivariate time series analysis and plant growth adaptability assessment.

[0173] Step S440: Call the plant growth adaptability assessment algorithm to perform multivariate time series analysis on the joint analysis dataset, calculate the correlation coefficient between the cold resistance gene expression product characteristics and environmental monitoring data within each time window, and determine the growth adaptability index for that time window based on the correlation coefficient.

[0174] Plant growth adaptability assessment algorithms are algorithms that comprehensively consider multiple factors to evaluate the plant's growth adaptability. Multivariate time series analysis analyzes time series data containing multiple variables to uncover the relationships and patterns of change among these variables.

[0175] In one implementation, step S440 may specifically include the following steps S441 - S444: Step S441: Divide the joint analysis dataset into multiple consecutive and non-overlapping time windows in chronological order.

[0176] A time window is a method for segmenting and analyzing time series data. When dividing a joint analysis dataset into multiple consecutive and non-overlapping time windows according to chronological order, it is necessary to determine the size of each time window. The size of the time window can be set according to actual needs and data characteristics; for example, it can be set to one day or one week.

[0177] For the joint analysis dataset, starting from the beginning of the time series, time windows are sequentially divided according to the set time window size to ensure that the data within each time window is continuous and that different time windows do not overlap. In this way, each time window contains environmental monitoring data and stress resistance trait characteristic sequences over a period of time, which facilitates local correlation analysis.

[0178] Step S442: Within each time window, calculate the Pearson correlation coefficient between the characteristics of cold-resistant gene expression products and various types of environmental monitoring data to obtain multiple single-factor correlation coefficients.

[0179] The Pearson correlation coefficient is a statistical indicator used to measure the linear correlation between two variables, with values ​​ranging from -1 to 1. Within each time window, Pearson correlation coefficients were calculated between the characteristics of cold-resistance gene expression products and various types of environmental monitoring data (such as temperature, humidity, and light intensity). These calculations yielded the univariate correlation coefficients between the characteristics of cold-resistance gene expression products and various types of environmental monitoring data within each time window. These coefficients reflect the degree of linear correlation between the characteristics of cold-resistance gene expression products and various types of environmental monitoring data within that time window.

[0180] Step S443: Construct a correlation coefficient matrix, and perform weighted summation of multiple single-factor correlation coefficients according to a preset weight allocation scheme to generate a comprehensive correlation coefficient.

[0181] The correlation coefficient matrix is ​​a matrix formed by combining multiple single-factor correlation coefficients within each time window. Each row of the matrix represents a time window, and each column represents the correlation coefficient between environmental monitoring data and the characteristics of cold-resistance gene expression products.

[0182] The preset weighting scheme is a set of weight values ​​pre-defined based on the importance of various types of environmental monitoring data to plant growth. Different types of environmental monitoring data may have varying degrees of impact on plant growth; therefore, it is necessary to assign an appropriate weight to each type of environmental monitoring data. Each single-factor correlation coefficient in the correlation coefficient matrix is ​​multiplied by its corresponding weight, and then all products within the same time window are summed to obtain the comprehensive correlation coefficient for that time window. The comprehensive correlation coefficient comprehensively considers the relationship between multiple environmental factors and the characteristics of cold-resistance gene expression products, thus more comprehensively reflecting the correlation between plant growth and the environment.

[0183] Step S444: Normalize the comprehensive correlation coefficient and convert it into a growth adaptability index.

[0184] Normalization transforms the comprehensive correlation coefficient to a predetermined range to facilitate comparison and evaluation. A possible normalization method is to transform the comprehensive correlation coefficient to the range [0, 1]. Through normalization, the comprehensive correlation coefficient is converted into a growth fitness index, allowing direct comparison of growth fitness across different time windows.

[0185] Step S450: Arrange the growth adaptability indicators of each time window in chronological order, generate the growth adaptability indicator change curve, and generate plant growth adaptability labels based on the trend characteristics of the growth adaptability indicator change curve.

[0186] The growth adaptability index change curve is formed by connecting the growth adaptability indices of different time windows in chronological order. This curve visually demonstrates how plant growth adaptability changes over time.

[0187] In one implementation, step S450 may specifically include the following steps S451 - S455: Step S451: Plot the growth adaptability index change curve with the time window number as the horizontal axis and the growth adaptability index value as the vertical axis.

[0188] When plotting the curve of changes in growth adaptability indicators, the time window number is used as the horizontal axis and the growth adaptability indicator value as the vertical axis. The coordinate points are connected sequentially according to the time series to form a curve. For example, plotting tools or programming languages ​​can be used to plot the curve. In Python, the Matplotlib library can be used to plot the curve of changes in growth adaptability indicators. By plotting the curve, the changing trend of plant growth adaptability over time can be observed intuitively.

[0189] Step S452: Perform trend analysis on the change curve, calculate the local trend line of the curve using the moving average algorithm, and identify the rising, falling and stable intervals of the trend line.

[0190] The moving average algorithm is used to smooth time series data and extract trend information. This algorithm smooths the curve by calculating the average value of data within a certain window, thereby reducing the impact of noise.

[0191] When performing trend analysis on the curves showing changes in growth adaptability indicators, choose an appropriate sliding window size, such as 3 or 5 time windows. For each point on the curve, calculate the average value of the data within its corresponding window to obtain the sliding average value for that point. Connect the sliding average values ​​of all points to form a local trend line. By observing the slope of the local trend line, the rising, falling, and stable intervals can be identified. The rising interval indicates that plant growth adaptability gradually increases over time, the falling interval indicates that growth adaptability gradually decreases, and the stable interval indicates that growth adaptability is relatively stable.

[0192] Step S453: Calculate the duration and magnitude of change of the index for each interval, where the magnitude of change is the difference between the index value at the start point and the index value at the end point of the interval.

[0193] For the identified rising, falling, and stable intervals, the duration and magnitude of indicator change for each interval are calculated. The duration refers to the number of time windows contained within the interval, reflecting the duration of the trend. The magnitude of indicator change is the difference between the growth fitness indicator value at the beginning and end of the interval. The magnitude of the change reflects the degree of change in plant growth fitness within that interval. By calculating the duration and magnitude of indicator change, a more detailed understanding of how plant growth fitness changes within different trend intervals can be obtained.

[0194] Step S454: Construct label description rules based on interval type, duration length, and change range.

[0195] The label description rules define plant growth adaptability labels based on interval type, duration, and magnitude of indicator change. For example, the following rules can be defined:

[0196] When the interval type is an upward interval, the duration is relatively long (e.g., exceeding a certain number of time windows) and the indicator changes significantly (e.g., exceeding a certain threshold), the label is "rapid increase in growth fitness". When the interval type is a downward interval, the duration is relatively long and the indicator changes significantly, the label is "rapid decrease in growth fitness". When the interval type is a stable interval, the duration is relatively long, the label is "stable growth fitness".

[0197] Based on actual needs and the characteristics of plant growth, more detailed and specific labeling rules can be developed to accurately describe different states of plant growth adaptability.

[0198] Step S455: Assign labels to the growth adaptability indicators of each time window in chronological order to generate plant growth adaptability labels.

[0199] After establishing the labeling rules, the growth adaptability indicators for each time window are labeled sequentially according to time. For each time window, the corresponding label is determined based on the interval type, the duration of the interval, and the magnitude of indicator change, according to the labeling rules.

[0200] The labels for each time window are combined to generate plant growth adaptation labels. These labels visually demonstrate the plant's growth adaptation over different time periods.

[0201] It is understood that the various algorithms involved in the above descriptions of the embodiments of the present invention can be obtained from relevant content in the prior art. To save space, they will not be elaborated on in the embodiments of the present invention. In addition, those skilled in the art can supplement the details based on common knowledge in the art when implementing the solutions of the present invention. For example, they can use normalization to eliminate dimensional conflicts before feature fusion, use interpolation to eliminate dimensional differences, reasonably set thresholds based on historical data, experience or business scenario requirements, train the model based on a general model training method, set the number of layers in the model structure based on actual needs, select activation functions, etc. The present invention will not provide redundant descriptions of overly detailed implementation processes here.

[0202] Please see Figure 2 , Figure 2This is a schematic diagram of a computer system provided in an embodiment of the present invention. The computer system includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 can be connected via a bus or other means. The processor 101 (or Central Processing Unit, CPU) is the computing and control core of the computer system, capable of parsing various instructions and processing various data within the computer system. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, mobile communication interface, etc.), and can be used to send and receive data under the control of the processor 101; the communication interface 102 can also be used for data transmission and interaction within the computer system. The memory 103 is a storage device in the computer system used to store programs and data. It is understood that the memory 103 here can include the computer system's built-in memory, or it can include extended memory supported by the computer system. The memory 103 provides storage space, which stores the computer system's operating system; this invention does not limit this storage space.

[0203] In one embodiment, the processor 101 executes the data labeling method for screening plants on steep slopes in high-altitude areas provided in the above embodiments of the present invention by running a computer program in the memory 103.

Claims

1. A data annotation method for screening plants on alpine steep slopes, characterized in that, The method includes: Multi-source collaborative data collection was carried out in the high-altitude and steep slope area, and topographic structure data, plant image data and environmental monitoring data of the high-altitude and steep slope area were acquired simultaneously. Spatiotemporal related datasets were constructed by aligning timestamps. A mapping model between vegetation indices and cold-resistant gene expression products is constructed based on a transfer learning framework. Vegetation index features are extracted from plant image data in the spatiotemporal associated dataset, and the mapping model is used to convert the vegetation index features into corresponding cold-resistant gene expression product features. A preset slope compensation model is invoked to perform terrain distortion correction processing on the terrain structure data in the spatiotemporal correlation dataset to obtain standardized terrain data that eliminates the influence of slope. The characteristics of the cold-resistant gene expression product are then calibrated using the standardized terrain data. Based on the spatial feature calibration of the cold-resistant gene expression product characteristics and the environmental monitoring data in the spatiotemporal correlation dataset, a time-series trend analysis is performed. The plant growth adaptability assessment algorithm is used to calculate the growth adaptability index for different time periods, and plant growth adaptability labels are generated based on the growth adaptability index.

2. The method according to claim 1, characterized in that, The method for constructing a mapping model between vegetation indices and cold-hardiness gene expression products based on a transfer learning framework involves extracting vegetation index features from plant image data in the spatiotemporally correlated dataset, and converting these vegetation index features into corresponding cold-hardiness gene expression product features using the mapping model. This includes: The plant image data in the spatiotemporal correlation dataset is subjected to spectral band separation processing to extract near-infrared band reflectance data and visible light band reflectance data, and initial vegetation index features are generated through a band combination algorithm. The initial vegetation index features are input into a pre-trained feature transfer network. The initial vegetation index features are then mapped to a high-dimensional feature space through the domain adaptation layer of the feature transfer network to generate a transfer feature vector. A preset dual-branch mapping model structure is obtained. The transfer feature vector is processed through the convolutional neural network architecture of the first branch to generate a convolutional feature map sequence. The temporal correlation model of the convolutional feature map sequence is performed through the recurrent neural network architecture of the second branch to generate a temporal feature vector. The temporal feature vector is input into the fusion output layer of the dual-branch mapping model structure. The contribution weights of convolutional features and temporal features are adjusted through an attention mechanism to generate a fusion feature vector. The fusion feature vector is then identified as the cold-resistant gene expression product feature, thus completing the feature transformation operation of the recessive stress resistance trait.

3. The method according to claim 2, characterized in that, The process involves performing spectral band separation processing on the plant image data in the spatiotemporal correlated dataset, extracting near-infrared and visible light reflectance data, and generating initial vegetation index features through a band combination algorithm, including: The plant image data is subjected to radiometric correction to generate a standardized spectral reflectance dataset; The standardized spectral reflectance dataset is subjected to band separation operation according to the preset band division threshold to obtain reflectance data sequences in the near-infrared band and reflectance data sequences in the visible light band. The near-infrared band reflectance data sequence and the visible light band reflectance data sequence are processed pixel by pixel band operation, and the initial vegetation index features are calculated by a preset band combination algorithm. The initial vegetation index features are spatially smoothed, and the median filtering algorithm is used to eliminate the interference of salt and pepper noise on the feature values, generating noise-suppressed initial vegetation index features. The dimensions of the initial vegetation index features after noise suppression are adjusted to match the input dimensions of the migration feature vector.

4. The method according to claim 1, characterized in that, The process involves calling a preset slope compensation model to correct terrain distortion in the spatiotemporal correlated dataset, resulting in standardized terrain data free from slope effects. This standardized terrain data is then used to perform spatial feature calibration on the cold-resistance gene expression product characteristics, including: Elevation and slope data are extracted from the topographic structure data in the spatiotemporal correlation dataset to construct a digital elevation model. The slope value distribution matrix is ​​then extracted from the digital elevation model using a slope calculation algorithm. The slope value distribution matrix is ​​input into the terrain factor calculation layer of the slope compensation model. The slope change rate and aspect variation coefficient are calculated by the terrain curvature analysis algorithm to generate the terrain distortion influence factor. Based on the terrain distortion influencing factors, a spatial correction function is constructed to perform geometric distortion correction on the digital elevation model, generating standardized terrain data that eliminates the influence of slope. Extract the spatial coordinate information of the cold-resistant gene expression product characteristics, perform spatial registration processing on the spatial coordinate information and the standardized terrain data, and establish a spatial correspondence between the cold-resistant gene expression product characteristics and the standardized terrain data; Based on the spatial correspondence, spatial interpolation is performed on the characteristics of the cold-resistant gene expression product, and the characteristic values ​​of the cold-resistant gene expression product are mapped to the grid cells of the standardized terrain data.

5. The method according to claim 4, characterized in that, The step of inputting the slope value distribution matrix into the terrain factor calculation layer of the slope compensation model, calculating the slope change rate and aspect variation coefficient through a terrain curvature analysis algorithm, and generating terrain distortion influence factors includes: Perform a first-order difference operation on the slope value distribution matrix to calculate the slope gradient values ​​in the horizontal and vertical directions, and construct the slope gradient matrix. The slope gradient matrix is ​​subjected to second-order differential processing using the terrain curvature analysis algorithm to calculate the profile curvature and planar curvature. The slope change rate is calculated based on the profile curvature and the plane curvature, whereby the slope change rate is a weighted sum of the absolute values ​​of the profile curvature and the plane curvature. The terrain structure data in the spatiotemporal correlation dataset is subjected to slope aspect extraction processing to generate a slope aspect angle matrix. The local standard deviation of the slope aspect angle matrix is ​​calculated, and the local standard deviation is determined as the slope aspect variation coefficient. The slope change rate and the aspect variation coefficient are input into a preset terrain distortion influence factor algorithm, and the two parameters are merged into a terrain distortion influence factor through normalization processing.

6. The method according to claim 5, characterized in that, The step of constructing a spatial correction function based on the terrain distortion influence factor to perform geometric distortion correction processing on the digital elevation model and generate standardized terrain data that eliminates the influence of slope includes: A spatial correction function is constructed with the topographic distortion influence factor as the independent variable, and the function value of the spatial correction function is positively correlated with the topographic distortion influence factor. For each elevation sampling point in the digital elevation model, a correction calculation is performed. The original elevation value of the corresponding elevation sampling point is multiplied by the function value of the spatial correction function at that sampling point location to obtain the preliminary corrected elevation value. The elevation value after preliminary correction is processed by a second correction based on the slope value of the sampling point to generate the elevation value after secondary correction. The digital elevation model is reconstructed using the secondary corrected elevation values ​​of all elevation sampling points to generate standardized terrain data that eliminates the influence of slope. The spatial resolution of the standardized terrain data is adjusted to match the spatial resolution of the cold-resistance gene expression product characteristics.

7. The method according to claim 1, characterized in that, The process involves performing time-series trend analysis on the cold-resistance gene expression product characteristics after spatial feature calibration and the environmental monitoring data in the spatiotemporal correlation dataset, calculating growth adaptability indicators for different time periods using a plant growth adaptability assessment algorithm, and generating plant growth adaptability labels based on these indicators. This includes: Environmental monitoring data is extracted from the spatiotemporal correlation dataset, and each type of environmental monitoring data is accompanied by timestamp information. The spatial feature calibration process is performed on the cold resistance gene expression product features. The cold resistance gene expression product feature values ​​at different collection times are extracted according to the preset time interval to generate a time-seriesd stress resistance trait feature sequence. The environmental monitoring data and the stress resistance trait sequence are aligned on the time axis to generate a spatiotemporally aligned joint analysis dataset; The plant growth adaptability assessment algorithm is called to perform multivariate time series analysis on the joint analysis dataset, calculate the correlation coefficient between the cold resistance gene expression product characteristics and environmental monitoring data in each time window, and determine the growth adaptability index of that time window based on the correlation coefficient. The growth adaptability indicators of each time window are arranged in chronological order to generate growth adaptability indicator change curves, and plant growth adaptability labels are generated based on the trend characteristics of the growth adaptability indicator change curves.

8. The method according to claim 7, characterized in that, The process involves extracting environmental monitoring data from the spatiotemporal correlation dataset. Each type of environmental monitoring data includes timestamp information, including: Data integrity checks are performed on the environmental monitoring data in the spatiotemporal correlation dataset to identify missing and abnormal data points. A linear interpolation algorithm is used to fill in the missing data points, and the 3σ criterion is used to replace the abnormal data points to generate a complete environmental monitoring dataset. The complete environmental monitoring dataset is classified and extracted according to the data collection type, resulting in different types of environmental datasets. Each dataset contains a timestamp field and a monitoring value field. Time format unification processing is performed on various types of environmental datasets, converting all timestamp information into a unified time encoding format; Time series resampling processing is performed on various types of environmental datasets. Interpolation or downsampling operations are performed on environmental monitoring data according to a preset time granularity to keep the time sampling interval of various types of environmental data consistent. The environmental monitoring data after resampling is standardized to convert the monitoring values ​​to a preset data range.

9. The method according to claim 8, characterized in that, The plant growth adaptability assessment algorithm is invoked to perform multivariate time series analysis on the joint analysis dataset, calculate the correlation coefficient between the cold-resistance gene expression product characteristics and environmental monitoring data within each time window, and determine the growth adaptability index for that time window based on the correlation coefficient, including: The joint analysis dataset is divided into multiple consecutive, non-overlapping time windows in chronological order; Within each time window, Pearson correlation coefficients were calculated between the characteristics of cold-resistance gene expression products and various types of environmental monitoring data to obtain multiple single-factor correlation coefficients; Construct a correlation coefficient matrix, and perform weighted summation of the multiple single-factor correlation coefficients according to a preset weight allocation scheme to generate a comprehensive correlation coefficient; The comprehensive correlation coefficient is normalized and converted into a growth adaptability index; The process of arranging growth adaptability indicators for each time window in chronological order to generate growth adaptability indicator change curves, and generating plant growth adaptability labels based on the trend characteristics of the growth adaptability indicator change curves, includes: Plot the growth adaptability index change curve with the time window number as the horizontal axis and the growth adaptability index value as the vertical axis. Trend analysis is performed on the change curve, and the local trend line of the curve is calculated by the moving average algorithm to identify the rising range, falling range and stable range of the trend line; Calculate the duration and magnitude of change of the index for each interval, where the magnitude of change is the difference between the index value at the start point and the index value at the end point of the interval. A label description rule is constructed based on the interval type, duration, and variation range. Plant growth adaptability labels are generated by assigning labels to the growth adaptability indicators of each time window in chronological order.

10. A computer system, characterized in that, include: A memory, wherein a computer program is stored; A processor is configured to load the computer program to implement the data labeling method for screening plants on alpine steep slopes as described in any one of claims 1-9.

Images