Intelligent alpine rhododendron identification method based on unmanned aerial vehicle sensor

Abstract

The invention discloses an intelligent alpine rhododendron identification method based on an unmanned aerial vehicle sensor, and relates to the technical field of image identification, and the method comprises the steps: obtaining an image sample set and an identification sample set of an alpine rhododendron planting region, carrying out the identification mode classification, and obtaining Q rhododendron identification mode regions; inputting the unmanned aerial vehicle remote sensing image sample set into a rhododendron lapponicum detection and segmentation module to obtain a rhododendron lapponicum image sample set; introducing a rhododendron identification learning channel to generate a rhododendron Q mode identification unit; generalization reinforcement learning is carried out on the Rhododendron Q mode identification unit; and obtaining a remote sensing acquisition image of the alpine rhododendron planting area, and generating a rhododendron Q mode identification result. The technical problems that in the prior art, the alpine rhododendron planting area identification accuracy is poor, the generalization ability is insufficient, and the alpine rhododendron planting area is difficult to adapt to multiple identification modes are solved, and the technical effects of achieving accurate identification of the alpine rhododendron planting area and improving the alpine rhododendron identification accuracy and generalization ability are achieved.

Description

Technical Field

[0001] The invention relates to the technical field of image recognition, and in particular to an intelligent recognition method for alpine rhododendron based on an unmanned aerial vehicle sensor. Background Art

[0002] In today's ecological protection and agricultural research fields, accurate identification and monitoring of alpine rhododendron planting areas are of vital importance. As a rare plant resource, alpine rhododendron not only plays an indispensable role in the ecosystem, but its planting conditions also have a profound impact on the maintenance of ecological balance and the development of the agricultural industry. However, there are many problems that need to be solved in the existing technologies for identifying alpine rhododendron planting areas. On the one hand, traditional recognition methods often rely on manual observation, which is not only inefficient, but also easily affected by the subjective factors of the observer, resulting in a significant reduction in the accuracy of the recognition results, making it difficult to meet large-scale, high-precision recognition needs. On the other hand, some existing automated recognition technologies have serious lack of generalization capabilities when faced with complex and changeable natural environments and different recognition modalities, and cannot effectively adapt to alpine rhododendron recognition tasks in different scenarios.

[0003] The existing technology has technical problems such as poor accuracy in identifying alpine rhododendron planting areas, insufficient generalization ability and difficulty in adapting to multiple recognition modalities. Summary of the invention

[0004] The present application provides an intelligent identification method for alpine rhododendrons based on drone sensors, which is used to solve the technical problems in the prior art of poor accuracy in identifying alpine rhododendron planting areas, insufficient generalization ability and difficulty in adapting to multiple recognition modes.

[0005] In view of the above problems, the present application provides an intelligent identification method of alpine rhododendron based on drone sensors, the method comprising: Acquire a UAV remote sensing image sample set and an alpine rhododendron recognition sample set of an alpine rhododendron planting area, and perform recognition modality classification on the alpine rhododendron recognition sample set to obtain Q rhododendron recognition modal areas, wherein Q is a positive integer greater than 1; input the UAV remote sensing image sample set into an alpine rhododendron detection and segmentation module to obtain an alpine rhododendron image sample set; introduce a rhododendron recognition learning channel, perform loss optimization learning on the alpine rhododendron image sample set and the Q rhododendron recognition modal areas, and generate a rhododendron Q modal recognition unit; introduce a GAN network and a predetermined remote sensing scene distribution comparison to perform generalized reinforcement learning on the rhododendron Q modal recognition unit to obtain a rhododendron modal recognition integrated module, and build an alpine rhododendron recognition module in combination with the alpine rhododendron detection and segmentation module; obtain a remote sensing acquisition image of the alpine rhododendron planting area according to a UAV sensor, input the remote sensing acquisition image into the alpine rhododendron recognition module, and generate a rhododendron Q modal recognition result.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: The UAV remote sensing image sample set and the alpine rhododendron identification sample set of the alpine rhododendron planting area are obtained, and the alpine rhododendron identification sample set is classified by identification modality to obtain Q rhododendron identification modal areas; the UAV remote sensing image sample set is input into the alpine rhododendron detection and segmentation module to obtain the alpine rhododendron image sample set; the alpine rhododendron image sample set and the Q rhododendron identification modal areas are subjected to loss optimization learning to generate the rhododendron Q modal identification unit; generalized reinforcement learning is performed to obtain the rhododendron modal identification integration module, and the alpine rhododendron identification module is constructed; the remote sensing acquisition image of the alpine rhododendron planting area is obtained according to the UAV sensor, and the image is input into the alpine rhododendron identification module to generate the rhododendron Q modal identification result. The technical effect of realizing accurate identification of the alpine rhododendron planting area and improving the accuracy and generalization ability of alpine rhododendron identification is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic diagram of a flow chart of an intelligent identification method for alpine azalea based on a drone sensor provided in an embodiment of the present application; Figure 2 A schematic diagram of the process of generating a cuckoo Q modal recognition unit in the intelligent recognition method of alpine cuckoo based on drone sensors provided in an embodiment of the present application. DETAILED DESCRIPTION

[0009] This application provides an intelligent identification method for alpine rhododendrons based on drone sensors, which is used to solve the technical problems in the prior art of poor accuracy in identifying alpine rhododendron planting areas, insufficient generalization ability and difficulty in adapting to multiple recognition modes.

[0010] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0011] Examples, such as Figure 1As shown, the present application provides an intelligent identification method of alpine rhododendron based on drone sensors, the method comprising: Step S100: Obtain a UAV remote sensing image sample set and an alpine rhododendron identification sample set of the alpine rhododendron planting area, and perform identification modality classification on the alpine rhododendron identification sample set to obtain Q rhododendron identification modal areas, where Q is a positive integer greater than 1.

[0012] Specifically, firstly, with the help of drones equipped with multispectral cameras, according to the topography, area size and distribution of the alpine rhododendron planting area, appropriate flight altitudes and routes are planned to obtain comprehensive and high-resolution multispectral images, forming a sample set of drone remote sensing images. At the same time, the variety, growth stage, leaf characteristics, flowering status and other information of the alpine rhododendron are recorded in detail to construct an alpine rhododendron identification sample set. Afterwards, the identification sample set is carefully classified from the dimensions of variety differences, growth process, leaf morphology and flowering degree, thereby obtaining multiple (Q, Q greater than 1) rhododendron identification modal areas with different characteristics. These modal areas cover various characteristics of alpine rhododendrons, laying a solid data foundation for the subsequent accurate identification of alpine rhododendrons, which will help improve the efficiency and accuracy of alpine rhododendron identification work.

[0013] Step S200: inputting the UAV remote sensing image sample set into the alpine rhododendron detection and segmentation module to obtain the alpine rhododendron image sample set.

[0014] Specifically, after obtaining the UAV remote sensing image sample set, this image sample set is input into the alpine rhododendron detection and segmentation module, which is composed of the preprocessing unit, the rhododendron pixel detection unit and the rhododendron background separation unit. First, the image is processed by the radiation correction model, the geometric correction model and the Gaussian filter in the preprocessing unit to correct the radiation error, geometric deformation and remove the noise, and then converted into a standard remote sensing image sample set. Then, the images are extracted one by one from the standard sample set, and one of them is taken as an example and input into the rhododendron pixel detection unit. The rhododendron pixel prediction model and the rhododendron pixel discrimination model in this unit cooperate. The prediction model first processes the pixel samples in the image to obtain the probability coefficient of each pixel belonging to the alpine rhododendron pixel, and then the discrimination model combines the predetermined coefficients to judge these prediction coefficients, and then determine the rhododendron pixel distribution. Finally, the rhododendron background separation unit performs background separation processing on the standard remote sensing image based on this pixel distribution, and successfully obtains a clear alpine rhododendron image sample set.

[0015] Step S300: introducing a cuckoo recognition learning channel, performing loss optimization learning on the alpine cuckoo image sample set and the Q cuckoo recognition modal areas, and generating a cuckoo Q modal recognition unit.

[0016] Specifically, after the acquisition of the alpine azalea image sample set, the azalea recognition learning channel is introduced, which contains P azalea recognition learners and a azalea recognition loss calculator (P is a positive integer greater than 1). For Q azalea recognition modal areas (such as modal areas formed by alpine azalea varieties, growth stages, etc.), the qth modal area (1≤q≤Q) is extracted from them each time. With the modal area and the alpine azalea image sample set as basic data, the azalea recognition loss calculator is used to supervise the training of the P azalea recognition learners. During the training process, each time the predetermined number of training times is reached, the P azalea recognition loss coefficients are calculated according to a specific azalea recognition loss calculation formula. If these loss coefficients are less than the set azalea recognition loss threshold, P same-modal azalea recognition models are generated, and they are connected as parallel nodes to form the qth modal azalea recognition model. Then, the output data set of the qth modal azalea recognition model is used as input, and the qth azalea recognition modal area is used as output to train the qth modal azalea recognition fusion model. Finally, the qth modal cuckoo recognition model is merged with the input layer of the fusion model to generate the qth modal recognition unit of cuckoo. The above process is repeated continuously to aggregate the generated qth modal recognition units to finally form the cuckoo Q modal recognition unit, thereby optimizing the model's recognition ability for alpine cuckoos in different modalities.

[0017] Step S400: introducing a GAN network and a predetermined remote sensing scene distribution comparison to perform generalized reinforcement learning on the cuckoo Q modal recognition unit to obtain a cuckoo modal recognition integrated module, and combining the alpine cuckoo detection and segmentation module to build an alpine cuckoo recognition module.

[0018] Specifically, after the generation of the cuckoo Q modal recognition unit, in order to improve the generalization ability of the model, the GAN network and the predetermined remote sensing scene distribution are introduced. First, by collecting the remote sensing scene information of the alpine azalea image sample set, the remote sensing scene distribution ratio is calculated, and compared with the predetermined remote sensing scene distribution ratio. The difference between the two is used as the optimization target to expand the remote sensing scene sample set. Based on the expanded sample set, the GAN network is used to perform data expansion on the alpine azalea image sample set to obtain the expanded azalea image sample set. Then, according to the intrinsic connection between the alpine azalea image sample set and the Q cuckoo recognition modal areas, a cuckoo image-recognition modal mapping network is established. With the help of this network, the expanded azalea image sample set is migrated and annotated according to the Q cuckoo recognition modal areas to generate an expanded azalea recognition modal sample set. Then, the expanded azalea image sample set and the expanded azalea recognition modal sample set are used to perform generalized reinforcement learning on the cuckoo Q modal recognition unit, thereby obtaining the cuckoo modal recognition integrated module. Finally, the rhododendron modal recognition integration module was connected with the previous alpine rhododendron detection and segmentation module to successfully build an alpine rhododendron recognition module, which enables the module to better adapt to different remote sensing scenarios and improve the recognition accuracy and stability of alpine rhododendrons.

[0019] Step S500: obtaining a remote sensing image of the alpine azalea planting area according to a drone sensor, inputting the remote sensing image into the alpine azalea recognition module, and generating a azalea Q modal recognition result.

[0020] Specifically, drones equipped with specific sensors were used to collect data from the alpine azalea planting areas again. The drones collected multispectral remote sensing images over the planting areas according to the established flight routes and parameters, and obtained remote sensing images containing rich information. These images were then input into the alpine azalea identification module that was carefully built in the early stage. The various components within the module work together to analyze and process the input images, identify and classify the alpine azaleas in the images through deep learning algorithms and previously trained model parameters, and finally generate the azalea Q modal recognition results, which can present in detail the variety, growth stage, leaf characteristics, flowering status and other information of the alpine azaleas in the image, providing key data support for the research, protection and monitoring of alpine azaleas.

[0021] In one possible implementation, Figure 2 As shown, step S300 also includes: Step S310: The cuckoo recognition learning channel includes P cuckoo recognition learners and a cuckoo recognition loss calculator, where P is a positive integer greater than 1.

[0022] Step S320: extracting the qth cuckoo identification modal area according to the Q cuckoo identification modal areas, where q is a positive integer, 1≤q≤Q.

[0023] Step S330: Based on the azalea recognition loss calculator, according to the alpine azalea image sample set and the qth azalea recognition modal area, loss optimization learning is performed on the P azalea recognition learners to generate a qth modal azalea recognition model.

[0024] Step S340: using the output data set of the qth modal cuckoo recognition model as input information and the qth cuckoo recognition modal area as output information to train the qth modal cuckoo recognition fusion model.

[0025] Step S350: Merge the qth modal cuckoo recognition model with the input layer of the qth modal cuckoo recognition fusion model to generate a cuckoo qth modal recognition unit, and add the cuckoo qth modal recognition unit to the cuckoo Q modal recognition unit.

[0026] Specifically, the cuckoo recognition learning channel is mainly composed of P cuckoo recognition learners and a cuckoo recognition loss calculator, where P represents a positive integer greater than 1. These P cuckoo recognition learners are all machine learning models, and they have powerful data learning and feature extraction capabilities. Different learners can analyze and learn the image data and recognition modal area data of alpine cuckoo from different angles. Some learners are good at capturing the texture features of images, while others are more sensitive to spectral features. The cuckoo recognition loss calculator is used to measure the deviation in the model learning process. In the subsequent learning process, the loss value of each learner in the learning process is calculated to evaluate the learner's fit to the data and the recognition effect, and provide a quantitative basis for optimizing the learner's parameters, so that the entire learning channel can efficiently learn and process the alpine cuckoo-related data, laying a solid foundation for generating more accurate recognition models.

[0027] After completing the modal classification of the alpine azalea identification sample set and obtaining Q azalea identification modal areas, select specific modal areas in order for subsequent processing. Among them, q represents a positive integer, and the value range is limited to between 1 and Q, that is, 1≤q≤Q. In actual operation, the qth cuckoo identification modal area is accurately extracted from the Q cuckoo identification modal areas. For example, if Q is 5, when the value of q is 3, the third modal area of ​​the five modal areas is extracted. This extracted qth cuckoo identification modal area will serve as the key data input for subsequent model training and optimization, providing basic data support for building a more targeted qth modal cuckoo identification model, thereby improving the recognition ability of alpine azaleas under different modalities.

[0028] The cuckoo recognition loss calculator performs supervised training on P cuckoo recognition learners. During the training process, P cuckoo recognition loss coefficients are calculated once every predetermined number of trainings. This coefficient is derived from the cuckoo recognition loss calculation formula, which compares the number of cuckoo recognition prediction samples and the number of correct prediction samples in each training, thereby measuring the degree of deviation between the model prediction and the actual situation. As the training continues, each learner will continuously adjust its own parameters according to the loss coefficient to find the optimal parameter combination that minimizes the loss. This is the process of loss optimization learning. When the P cuckoo recognition loss coefficients are all less than the pre-set cuckoo recognition loss threshold, it indicates that the training of the learner has reached a certain accuracy requirement. At this time, these P well-performing isomodal cuckoo recognition models are connected as parallel nodes to generate the qth modal cuckoo recognition model, which can more effectively identify alpine cuckoos belonging to the qth modality.

[0029] The output data set of the qth modal cuckoo recognition model is used as the new input information, because the output data set already contains the model's preliminary extraction and analysis results of the alpine cuckoo image features. At the same time, the qth cuckoo recognition modal area is used as the expected output information. This modal area covers the accurate features and classification information of alpine cuckoo in a specific mode. Using these two sets of data, the back propagation algorithm is used to train the qth modal cuckoo recognition fusion model. During the training process, the fusion model continuously adjusts its own network parameters to try to minimize the difference between the predicted output and the actual qth cuckoo recognition modal area. Through multiple iterative training, the fusion model gradually learns how to better integrate and process the output information of the qth modal cuckoo recognition model to more accurately match the qth cuckoo recognition modal area, thereby improving the recognition accuracy and stability of the qth modal alpine cuckoo.

[0030] After completing the training of the qth modal cuckoo recognition model and the qth modal cuckoo recognition fusion model, the input layer of the qth modal cuckoo recognition model and the qth modal cuckoo recognition fusion model are merged. Through the neural network structure fusion algorithm, the neuron connection relationship of the input layers of the two models is reintegrated, so that the newly generated structure can simultaneously receive feature information from the input data of the two models, thereby forming a more powerful cuckoo qth modal recognition unit. This new unit combines the ability of the qth modal cuckoo recognition model to initially extract features and the advantages of the qth modal cuckoo recognition fusion model to optimize and accurately match features. After the cuckoo qth modal recognition unit is generated, it is added to the cuckoo Q modal recognition unit according to the established data storage and management rules. As this operation is performed for each q value (1≤q≤Q), the cuckoo Q modal recognition unit is continuously improved, which covers the recognition capabilities of alpine cuckoos for different modalities, and provides the core technical support for the subsequent accurate recognition of alpine cuckoos.

[0031] In a possible implementation, step S330 further includes: Step S331: According to the alpine rhododendron image sample set and the qth rhododendron recognition modal area, supervised training is performed on the P rhododendron recognition learners respectively, and each time a predetermined number of training times is trained, P rhododendron recognition loss coefficients are calculated according to the rhododendron recognition loss calculator, wherein the rhododendron recognition loss calculator includes a rhododendron recognition loss calculation formula, and the rhododendron recognition loss calculation formula is: ; Among them, LOSS represents the cuckoo recognition loss coefficient, M represents the scheduled number of training times, m represents the m-th training, M and m are both positive integers, 1≤m≤M, SUOm represents the number of cuckoo recognition prediction samples in the m-th training, and SUXm represents the number of correct samples of cuckoo recognition prediction in the m-th training.

[0032] Step S332: If the P cuckoo recognition loss coefficients are less than the cuckoo recognition loss threshold, generate P same-modal cuckoo recognition models.

[0033] Step S333: connecting the P same-modal cuckoo recognition models as parallel nodes to generate the q-th modal cuckoo recognition model.

[0034] Specifically, supervised training of P cuckoo recognition learners. Taking the commonly used deep learning framework TensorFlow as an example, the data is first preprocessed, and the data of the alpine rhododendron image sample set and the qth cuckoo recognition modal area are converted into a format acceptable to the model. The image sample set is normalized so that the pixel value is between 0-1, and the data of the qth cuckoo recognition modal area is encoded to correspond to the image data. During the training process, a suitable predetermined number of training times is set, assuming it is 50 times. During each training, the data is input into P cuckoo recognition learners in batches, and each learner attempts to recognize the alpine rhododendron image samples and predict the modal category to which it belongs. After each training is completed, relevant data will be recorded, including the number of cuckoo recognition prediction samples SUOm and the number of predicted correct samples SUXm in the mth training. When the training reaches the predetermined number of training times, the cuckoo recognition loss calculator comes into play. It calculates the cuckoo recognition loss based on the given formula , calculate the loss coefficients of the P cuckoo recognition learners respectively. In this formula, M represents the predetermined number of training times (here 50), m represents the current number of training (1≤m≤M), and LOSS is the calculated cuckoo recognition loss coefficient, which reflects the degree of deviation between the model prediction result and the actual situation. After the calculation is completed, the obtained P cuckoo recognition loss coefficients are compared with the preset cuckoo recognition loss threshold. If all the loss coefficients are less than the threshold, it means that the P cuckoo recognition learners perform well under the current training, and the learning of the alpine cuckoo image and the qth cuckoo recognition modal area has reached a certain accuracy requirement. At this time, the models corresponding to the P learners with excellent performance are saved to generate P same-modal cuckoo recognition models. Finally, in order to further improve the recognition ability of the model, these P same-modal cuckoo recognition models are connected as parallel nodes. In actual operation, the input layer and output layer of these models are reasonably connected using the functions or tools in TensorFlow so that they can process and analyze the input data at the same time. After such processing, a q-th modal cuckoo recognition model with stronger comprehensive performance was generated. It integrates the advantages of P same-modal cuckoo recognition models and can more accurately identify the alpine cuckoo belonging to the q-th mode, providing key support for subsequent identification work.

[0035] In a possible implementation, step S400 further includes: Step S410: Based on the GAN network, data expansion is performed on the alpine rhododendron image sample set according to the predetermined remote sensing scene distribution ratio to obtain an expanded rhododendron image sample set.

[0036] Step S420: establishing a cuckoo image-recognition modality mapping network based on the mapping relationship between the alpine azalea image sample set and the Q cuckoo recognition modality areas.

[0037] Step S430: Based on the cuckoo image-recognition modality mapping network, the expanded cuckoo image sample set is migrated and labeled according to the Q cuckoo recognition modality areas to generate an expanded cuckoo recognition modality sample set.

[0038] Step S440: performing generalized reinforcement learning on the cuckoo Q modality recognition unit according to the expanded cuckoo image sample set and the expanded cuckoo recognition modality sample set to generate the cuckoo modality recognition integrated module.

[0039] Step S450: connecting the azalea modal recognition integration module and the alpine azalea detection and segmentation module to obtain the alpine azalea recognition module.

[0040] Specifically, the data expansion work is carried out using the generative adversarial network (GAN network), and the predetermined remote sensing scene distribution ratio is used as a reference and input into the GAN network. The GAN network consists of a generator and a discriminator. The generator is responsible for generating new simulated alpine rhododendron image data based on the predetermined remote sensing scene distribution ratio and the characteristics of the alpine rhododendron image sample set; the discriminator identifies the generated image data to determine whether it is real. Through the continuous game between the generator and the discriminator, the data generated by the generator is closer and closer to the real alpine rhododendron image, thereby realizing the data expansion of the alpine rhododendron image sample set, and finally obtaining a large number of expanded rhododendron image sample sets, enriching the diversity and quantity of image data.

[0041] Establishing a rhododendron image-recognition modality mapping network is a key link to lay the foundation for subsequent precise labeling and recognition work. First, deep feature extraction is performed on the sample set of alpine rhododendron images. Using the convolutional neural network (CNN), with its powerful convolution layer and pooling layer structure, multi-scale and multi-level feature extraction is performed on the image samples. The convolution layer slides on the image through different convolution kernels to extract local features in the image, such as edges and textures; the pooling layer performs dimensionality reduction processing on the output of the convolution layer to reduce the amount of data while retaining key features. After multiple layers of convolution and pooling operations, a feature vector that can effectively characterize the alpine rhododendron image is obtained. Then, the data of Q rhododendron recognition modal areas are encoded. According to different modal categories, such as alpine rhododendron varieties, growth stages, leaf characteristics, flowering status, etc., the data of each modal area is converted into a numerical vector representation. For example, for alpine rhododendron varieties, one-hot encoding can be used to represent each variety with a unique binary vector; for continuous data such as growth stages and leaf characteristics, normalization is performed to make its value within a certain reasonable range. Then, the extracted alpine rhododendron image feature vectors were correlated with the encoding vectors of Q rhododendron identification modal areas. A mapping model was constructed using the fully connected neural network (FCN) in deep learning, with the alpine rhododendron image feature vector as input and the encoding vectors of Q rhododendron identification modal areas as output. During the training process, the back propagation algorithm was used to continuously adjust the weights and biases of the fully connected layer so that the model could learn the intrinsic mapping relationship between the alpine rhododendron image features and the identification modal area. Through multiple iterative training, the error between the model's prediction results and the actual identification modal area encoding vectors was gradually reduced until the predetermined accuracy requirements were met, thereby successfully establishing a rhododendron image-identification modal mapping network, which can accurately map the features of the alpine rhododendron image to the corresponding identification modal area, providing strong support for subsequent migration and annotation work.

[0042] With the help of the established rhododendron image-recognition modality mapping network, the expanded rhododendron image sample set is migrated and annotated. Each image in the expanded rhododendron image sample set is input into the mapping network. According to the mapping relationship learned by the network and the information of Q rhododendron recognition modality areas, these expanded images are annotated to determine their corresponding rhododendron recognition modalities. For example, for a newly generated alpine rhododendron image, through the analysis of the mapping network, it is judged that it belongs to a certain alpine rhododendron variety, is in a specific growth stage, etc., thereby generating an expanded rhododendron recognition modality sample set, providing richer and more targeted annotation data for subsequent model training.

[0043] The expanded rhododendron image sample set and the expanded rhododendron recognition modal sample set are used to perform generalized reinforcement learning on the rhododendron Q modal recognition unit. These two sets of sample sets are input into the rhododendron Q modal recognition unit. By continuously adjusting the parameters of the model, the model can learn more alpine rhododendron features and recognition modal information in different scenarios, thereby improving the model's adaptability and generalization capabilities to different situations. During the training process, optimization algorithms such as gradient descent are used to continuously reduce the error between the model's prediction results and the actual annotations, so that the model can more accurately identify the modal information of various alpine rhododendrons, and finally generate a rhododendron modal recognition integrated module, which improves the recognition performance of the model.

[0044] In order to build a complete alpine azalea recognition module, it is necessary to connect the azalea modal recognition integration module and the alpine azalea detection and segmentation module. Before starting the connection work, it is necessary to ensure that the two modules have completed training and have stable performance. The alpine azalea detection and segmentation module uses the Canny edge detection algorithm as the core technology to process the original remote sensing image. The Canny edge detection algorithm is a classic edge detection algorithm. It uses a multi-step processing process to accurately extract the edge of the image. First, the input original remote sensing image is Gaussian filtered to remove noise interference in the image, make the image smoother, and avoid the influence of noise on subsequent edge detection. Then, the finite difference of the first-order partial derivative is used to calculate the gradient amplitude and direction of the image, so as to determine the areas with more drastic grayscale changes in the image, which often correspond to the edges of objects. Then, the non-maximum suppression technology is used to refine the gradient amplitude image, retain only the points with the largest local gradient as edge points, and suppress non-edge candidate points, so that the edges are clearer and more accurate. Finally, the double threshold algorithm is used to detect and connect the edges, connect the real edge points, and form a complete edge contour. After being processed by the Canny edge detection algorithm, the alpine azalea detection and segmentation module can accurately detect the position of the alpine azalea in the image, and segment the alpine azalea from the complex background environment based on the extracted edge contour, generating pure image data containing only alpine azalea. The azalea modal recognition integration module is responsible for in-depth analysis of the alpine azalea image data after detection and segmentation, and identifies various modal information of alpine azalea, such as variety, growth stage, and health status, based on its own algorithms and models. In the connection link, the output port of the alpine azalea detection and segmentation module is accurately connected to the input port of the azalea modal recognition integration module. When a new remote sensing image is input into the entire system, the image will first enter the alpine azalea detection and segmentation module. The module performs preliminary processing operations on the image according to the processing flow of the Canny edge detection algorithm, and extracts effective image information about the alpine azalea from it. The processed image data will be automatically transmitted to the azalea modal recognition integration module according to the established data transmission path. After receiving the data, the Rhododendron modal recognition integration module uses its internal algorithms and models to identify and judge the modal information of the alpine rhododendron. Through this connection method, the functions of the two modules are organically integrated to build a complete and coherent processing flow, and finally obtain the alpine rhododendron recognition module with detection, segmentation and modal recognition functions. This module can process the input remote sensing images in an integrated manner and directly output the recognition results covering all-round information such as the location, morphology and modality of the alpine rhododendron.

[0045] In a possible implementation, step S410 further includes: Step S411: collecting remote sensing scene information of the Rhododendron azalea image sample set to obtain a remote sensing scene sample set.

[0046] Step S412: Calculate the remote sensing scene distribution ratio according to the remote sensing scene sample set.

[0047] Step S413: setting a remote sensing scene distribution optimization target according to a gap between the predetermined remote sensing scene distribution ratio and the remote sensing scene distribution ratio.

[0048] Step S414: Based on the remote sensing scene distribution optimization target, the remote sensing scene sample set is expanded to obtain an expanded remote sensing scene sample set.

[0049] Step S415: Based on the expanded remote sensing scene sample set, data expansion is performed on the alpine rhododendron image sample set according to the GAN network to obtain the expanded rhododendron image sample set.

[0050] Specifically, remote sensing scene information of the alpine rhododendron image sample set is collected to obtain the remote sensing scene sample set. With the help of professional remote sensing data acquisition equipment and related technical means, comprehensive and detailed collection of various scene information involved in the alpine rhododendron image sample set is carried out. The collection content covers multiple dimensions of the growth environment of alpine rhododendron, including but not limited to topographic information, such as different terrain types such as mountains, plains, and river valleys; climate condition information, such as temperature, humidity, light intensity, rainfall, etc.; vegetation coverage information, including the types and density of other surrounding plants; and geographical location information, such as longitude, latitude, altitude, etc. Through precise acquisition equipment, these scene information are monitored and recorded in real time and organized into a standardized data format. After the acquisition is completed, the large amount of raw data obtained is screened, classified and organized, and duplicate or invalid data is eliminated to ensure the accuracy and reliability of the data, and finally a remote sensing scene sample set with a clear structure and complete content is formed, laying a solid data foundation for the subsequent in-depth analysis and processing of alpine rhododendron image data.

[0051] To calculate the remote sensing scene distribution ratio, we must first sort out and clarify the various scene information dimensions covered by the remote sensing scene sample set, including topography, climate conditions, vegetation coverage and other aspects. For example, the topography dimension involves different types such as mountains, plains, hills, and basins; the climate condition dimension covers different weather conditions such as sunny days, cloudy days, rainy days, and snowy days; the vegetation coverage dimension can be divided into different intervals according to the vegetation coverage rate, such as 0%~20%, 20~40%, etc., and statistical work is carried out for the specific information of each dimension. Taking the topography dimension as an example, it is necessary to count the number of times each terrain type appears in the sample set. By manually counting or using data analysis software, the entire remote sensing scene sample set is traversed to record the number of samples belonging to mountainous terrain. Similarly, the number of samples of other terrain types such as plains, hills, and basins is counted. After completing the statistics of the number of samples of all terrain types, calculate the total number of samples under this dimension, that is, add up the number of samples of all terrain types. Then, by dividing the number of samples of each terrain type by the total number of samples, we can get the proportion of each terrain type in the sample set, that is, the distribution ratio of the topography. The information of other dimensions is processed according to the same process. For the climate condition dimension, the number of samples corresponding to different weather conditions such as sunny days, cloudy days, rainy days, and snowy days is counted, and then the number of samples of each weather condition is divided by the total number of samples to obtain the distribution ratio of climate conditions. In the vegetation coverage dimension, the number of samples corresponding to different vegetation coverage intervals is counted, and the number of samples in each interval is divided by the total number of samples to obtain the distribution ratio of vegetation coverage. In this way, for each scene information dimension in the remote sensing scene sample set, the distribution ratio of different types under each dimension is calculated respectively, so as to obtain the remote sensing scene distribution ratio in a comprehensive and detailed manner. These distribution ratio data can intuitively and clearly show the proportion of various scene information in the sample set, providing a key and reference basis for subsequent data analysis, comparison and optimization adjustment.

[0052] Compare and analyze the pre-set remote sensing scene distribution ratio with the remote sensing scene distribution ratio just calculated. By calculating the difference between the two, the difference between the current sample set and the expected target in scene distribution is clarified, and this difference is used as the remote sensing scene distribution optimization target. This target will guide the subsequent sample expansion work to ensure that the expanded sample set is more in line with the predetermined requirements in scene distribution.

[0053] According to the determined remote sensing scene distribution optimization target, the remote sensing scene sample set is expanded in a targeted manner. By analyzing the deficiencies in the scene distribution in the current sample set, the missing or under-proportioned scene information is selectively supplemented. For example, if it is found that the scene information of a certain terrain and landform accounts for a small proportion in the sample set, the number of samples of this type of scene is increased through further data collection or simulation generation, thereby obtaining an expanded remote sensing scene sample set. The expanded sample set is more balanced in scene distribution and can better meet the needs of subsequent data processing.

[0054] With the help of the expanded remote sensing scene sample set, the generative adversarial network (GAN network) is used to expand the data of the alpine rhododendron image sample set. The GAN network consists of a generator and a discriminator. The generator generates new simulated alpine rhododendron image data based on the characteristics and distribution of the expanded remote sensing scene sample set; the discriminator identifies the generated image data to determine whether it is real. Through the continuous game between the generator and the discriminator, the data generated by the generator is closer and closer to the real alpine rhododendron image, and finally a large number of expanded rhododendron image sample sets are obtained, providing richer and more diverse image data support for subsequent model training and recognition work.

[0055] In a possible implementation, step S200 further includes: Step S210: the alpine rhododendron detection and segmentation module includes a pre-processing unit, a rhododendron pixel detection unit and a rhododendron background separation unit.

[0056] Step S220: inputting the UAV remote sensing image sample set into the preprocessing unit to obtain a standard remote sensing image sample set, wherein the preprocessing unit includes a radiation correction model, a geometric correction model and a Gaussian filter.

[0057] Step S230: traverse the standard remote sensing image sample set to extract the first standard remote sensing image sample.

[0058] Step S240: inputting the first standard remote sensing image sample into the azalea pixel detection unit to obtain the azalea pixel distribution of the first image.

[0059] Step S250: Based on the rhododendron pixel distribution of the first image, the rhododendron background separation unit performs background separation processing on the first standard remote sensing image sample to obtain a first alpine rhododendron image sample.

[0060] Step S260: adding the first Rhododendron azalea image sample to the Rhododendron azalea image sample set.

[0061] Specifically, the alpine rhododendron detection and segmentation module is composed of three sub-units with different functions but closely coordinated: the preprocessing unit, the rhododendron pixel detection unit, and the rhododendron background separation unit. The preprocessing unit is the starting point of the entire detection and segmentation process and is responsible for improving the quality of the input image. It is equipped with professional tools such as the radiation correction model, the geometric correction model, and the Gaussian filter. The radiation correction model can effectively eliminate the image radiation error caused by factors such as sensor performance differences, atmospheric scattering and absorption, and changes in lighting conditions, ensuring that the image brightness, color and other information can truly reflect the actual scene; the geometric correction model focuses on correcting the image geometric deformation caused by factors such as unstable drone flight posture, undulating terrain, and shooting angles, so that the position and shape of the objects in the image are restored to an accurate state; the Gaussian filter is mainly used to smooth the image, effectively filter out the noise introduced during the image acquisition process, and improve the clarity and recognizability of the image. The rhododendron pixel detection unit performs a deep analysis of the preprocessed image, and determines whether each pixel belongs to the alpine rhododendron pixel by pixel, thereby accurately determining the pixel distribution of the alpine rhododendron in the image. The rhododendron background separation unit accurately distinguishes the alpine rhododendron from the complex background environment based on the pixel distribution information provided by the rhododendron pixel detection unit, removes the background part, and retains only the pure alpine rhododendron image information, providing high-quality image data for subsequent alpine rhododendron identification work.

[0062] The UAV remote sensing image sample set is sent to the preprocessing unit to obtain the standard remote sensing image sample set. The preprocessing unit is a key link to improve image quality, which includes a variety of processing models and filters. Among them, the radiation correction model corrects the radiation brightness of the image through the formula L=(DN-b)×K (L is the radiation brightness, DN is the digital quantization value of each pixel, b is the bias value, and K is the gain coefficient), eliminates the radiation error caused by sensor characteristics, atmospheric conditions and other factors, and makes the image more realistic. The geometric correction model is used to correct the geometric deformation of the image caused by the flight posture of the UAV, terrain undulations, etc., so that the position of the ground objects in the image is more accurate; the Gaussian filter can smooth the image, remove noise interference, and improve image quality. After this series of processing, the UAV remote sensing image sample set is converted into a standard remote sensing image sample set, providing a high-quality data foundation for subsequent detection and segmentation of alpine rhododendrons.

[0063] The standard remote sensing image sample set is traversed, and during the traversal process, the image samples are extracted one by one, and the currently extracted sample is named the first standard remote sensing image sample. This operation is to process each image in the sample set separately to ensure that the subsequent detection and separation work can be accurately applied to each image sample.

[0064] In order to obtain the pixel distribution of rhododendron in the first image, the first standard remote sensing image sample needs to be input into the rhododendron pixel detection unit. The following is the specific implementation process. The first is data preparation. It is necessary to ensure that the format of the first standard remote sensing image sample can be effectively processed by the detection unit and convert the image into a specific data format. For example, in common image analysis, the image data will be converted into a suitable array form, which is convenient for subsequent feature extraction and model processing. Next, enter the feature extraction stage, in which image features are extracted from multiple dimensions. On the one hand, color feature extraction is color feature extraction. Different types of alpine rhododendrons have unique colors, such as bright flowers and relatively fixed leaf colors. In order to better capture these color information, the image will be converted from the common RGB color space to the HSV color space. By setting a specific color range, the pixel area that matches the color characteristics of the alpine rhododendron can be screened out. For example, if the alpine rhododendron flower is red, the red color threshold range can be set, and the corresponding color mask can be generated using image processing technology to mark the pixel area that may belong to the alpine rhododendron. On the other hand, texture feature extraction is used. The leaves and flowers of alpine azalea have unique textures. The gray-level co-occurrence matrix (GLCM) can be used to effectively extract the texture features of the image. First, the image is converted into a grayscale image, and then the gray-level co-occurrence matrix is ​​calculated. Through this matrix, texture features such as contrast and correlation can be further extracted. These texture features help to more accurately distinguish alpine azalea pixels from other background pixels. Then comes the model prediction stage, using a pre-trained pixel classification model, such as a convolutional neural network (CNN) for prediction. During the training process, the model learns a large number of alpine azalea image features and corresponding pixel classification labels, so it can perform pixel classification prediction based on the input first standard remote sensing image sample. Before inputting the image, the image needs to be adjusted to the size required by the model to ensure that the model can process normally. After being processed by the model, the probability of each pixel belonging to different categories will be output. By selecting the category with the highest probability as the classification result of the pixel, the pixel classification of the entire image is obtained. Finally, the rhododendron pixel distribution stage is determined. According to the pixel classification results predicted by the model, it is determined which pixels belong to alpine azalea. Alpine azalea pixels can be screened out by setting a specific threshold or based on category labels. After the screening is completed, there will be some small isolated pixel areas, which may be the result of noise or misclassification. In order to make the cuckoo pixel distribution more accurate and clear, morphological processing methods are used, such as removing small isolated areas, and finally obtaining the accurate cuckoo pixel distribution of the first image. This distribution information provides a key basis for subsequent background separation and other processing.

[0065] The GrabCut algorithm is used to perform background separation on the first standard remote sensing image sample in combination with the pixel distribution of rhododendron in the first image to obtain the first alpine rhododendron image sample. First, an initial mask is constructed based on the pixel distribution of rhododendron in the first image, and the pixels determined to be alpine rhododendron are marked as foreground, the pixels determined to be background are marked as background, and the uncertain pixels are marked as unknown areas. Then, the GrabCut algorithm is started with this mask and the first standard remote sensing image sample as input. The algorithm models the color distribution of the foreground and background based on the Gaussian mixture model (GMM) and continuously updates the pixel labels of the foreground and background through iterative optimization. In each iteration, the algorithm re-estimates the parameters of the GMM based on the current pixel labels, and then reclassifies the pixels in the unknown area based on the new GMM to determine whether they are more likely to belong to the foreground or the background. After multiple iterations, until the pixel labels no longer change significantly or the preset number of iterations is reached, finally, the first alpine rhododendron image sample is extracted from the first standard remote sensing image sample based on the determined foreground pixels to achieve background separation.

[0066] The first alpine rhododendron image sample obtained after background separation is added to the alpine rhododendron image sample set. This operation continuously enriches the content of the alpine rhododendron image sample set, accumulates more data resources for subsequent model training, data analysis and other work, and helps to improve the performance and accuracy of the entire alpine rhododendron recognition.

[0067] In a possible implementation, step S240 further includes: Step S241: the cuckoo pixel detection unit includes a cuckoo pixel prediction model and a cuckoo pixel discrimination model.

[0068] Step S242: obtaining a plurality of pixel samples of the first standard remote sensing image sample.

[0069] Step S243: inputting the plurality of pixel samples into the azalea pixel prediction model to obtain a plurality of azalea pixel prediction coefficients, wherein each azalea pixel prediction coefficient is used to characterize the probability that each pixel sample belongs to an alpine azalea pixel.

[0070] Step S244: inputting the plurality of cuckoo pixel prediction coefficients into the cuckoo pixel discrimination model to obtain a plurality of cuckoo pixel discrimination results, wherein the cuckoo pixel discrimination model includes predetermined cuckoo pixel prediction coefficients.

[0071] Step S245: extracting cuckoo pixels from the first standard remote sensing image sample based on the multiple cuckoo pixel identification results to generate the first image cuckoo pixel distribution.

[0072] Specifically, in the alpine rhododendron detection and segmentation module, the rhododendron pixel prediction model in the rhododendron pixel detection unit adopts the convolutional neural network (CNN) algorithm. First, a large number of remote sensing images containing alpine rhododendrons are collected and pixel categories are annotated, and the training set, validation set and test set are divided; a CNN with a U-Net architecture is constructed, and the convolution layer is used to extract local pixel features, the pooling layer is used to reduce the dimension, and the deconvolution layer is used to restore the size; the model is trained with the cross entropy loss function and the Adam optimizer, and the validation set is used to prevent overfitting; multiple pixel samples of the first standard remote sensing image sample are input into the trained model, and multiple rhododendron pixel prediction coefficients are obtained through forward propagation. The rhododendron pixel discrimination model adopts a threshold-based discrimination algorithm. According to the training set and validation set data, the predetermined rhododendron pixel prediction coefficient is determined through experiments or optimization methods, and multiple rhododendron pixel prediction coefficients are compared with it. The corresponding pixel samples greater than or equal to the threshold are judged as alpine rhododendron pixels, and those less than the threshold are judged as non-alpine rhododendron pixels, so as to achieve accurate discrimination of pixel samples.

[0073] To further analyze the first standard remote sensing image sample to detect the alpine rhododendron pixels, it is necessary to obtain multiple pixel samples of the image sample. First, considering that the image itself is a two-dimensional matrix structure composed of many pixels arranged in specific rows and columns, each row and column of the image is directly traversed, and each pixel is extracted as a separate sample. For color images, each pixel is usually represented by the values ​​of the three channels of red (R), green (G), and blue (B), so when extracting pixel samples, the specific values ​​of these three channels will be obtained at the same time. If there is additional band information in the image, such as the near-infrared band, the pixel values ​​of these bands will also be extracted. In the extraction process, the strategy of block sampling is also adopted to divide the first standard remote sensing image sample into multiple sub-blocks of the same size, and select some pixels from each sub-block as samples, which can not only ensure the acquisition of pixel features in different regions, but also reduce the amount of calculation to a certain extent, and improve the efficiency of subsequent analysis. Finally, multiple pixel samples are obtained in the above manner to provide basic data for subsequent rhododendron pixel prediction and discrimination.

[0074] After obtaining multiple pixel samples of the first standard remote sensing image sample, these samples are input into the rhododendron pixel prediction model. The model is based on deep learning algorithms, such as convolutional neural networks (CNNs), and is trained with a large amount of labeled alpine rhododendron image data. Inside the model, each pixel sample first passes through a series of convolutional layers, and the convolution kernels in the convolutional layers extract features from the pixel samples to capture their color, texture and other feature information. Next, the feature map is downsampled through the pooling layer to reduce the amount of data and enhance the robustness of the features. Subsequently, the extracted features are integrated and mapped through the fully connected layer. Finally, the model uses the softmax activation function to output the rhododendron pixel prediction coefficient corresponding to each pixel sample. The value range of this coefficient is between 0 and 1, which is used to accurately characterize the probability that the pixel sample belongs to an alpine rhododendron pixel, providing a key basis for the subsequent judgment of whether the pixel is an alpine rhododendron pixel.

[0075] After obtaining multiple rhododendron pixel prediction coefficients, they are input into the rhododendron pixel discrimination model. The model has a built-in predetermined rhododendron pixel prediction coefficient, which is a key threshold determined after training and verification of a large number of alpine rhododendron image samples. The model will compare each rhododendron pixel prediction coefficient with the predetermined rhododendron pixel prediction coefficient in turn. When a certain rhododendron pixel prediction coefficient is greater than or equal to the predetermined rhododendron pixel prediction coefficient, it indicates that the possibility that the pixel sample corresponding to the coefficient belongs to the alpine rhododendron pixel reaches or exceeds the preset standard. At this time, the pixel sample is marked as an alpine rhododendron pixel; and when the rhododendron pixel prediction coefficient is less than the predetermined rhododendron pixel prediction coefficient, it means that the possibility that the pixel sample belongs to the alpine rhododendron pixel does not meet the preset standard, so it is marked as a non-alpine rhododendron pixel. Through such comparison and marking operations, the model finally outputs multiple rhododendron pixel discrimination results, which provides a clear basis for the subsequent accurate extraction of alpine rhododendron pixels from the first standard remote sensing image sample.

[0076] Based on the accurate extraction of rhododendron pixels on the first standard remote sensing image sample, the first image rhododendron pixel distribution is generated. At this time, the entire first standard remote sensing image sample is traversed, and each pixel is matched one by one with the previously obtained rhododendron pixel discrimination result. For those pixels marked as alpine rhododendron pixels in the discrimination result, they are selected from the image sample, and the pixels related to alpine rhododendron are retained. The pixels marked as non-alpine rhododendron pixels will be ignored. With the completion of the traversal, all the alpine rhododendron pixels that have been screened out constitute a new set, and the position and distribution information of these pixels in the new set is the first image rhododendron pixel distribution. This distribution information intuitively shows the specific position of the alpine rhododendron pixels in the first standard remote sensing image sample, clearly presents the outline and morphology of the alpine rhododendron in the image, and further improves the accuracy and efficiency of the alpine rhododendron image analysis.

[0077] In a possible implementation, step S250 further includes: Step S251: performing contour recognition according to the first alpine rhododendron image sample to obtain a first rhododendron contour recognition result.

[0078] Step S252: performing a smoothness evaluation according to the first cuckoo contour recognition result to obtain the first cuckoo contour smoothness.

[0079] Step S253: determining whether the first cuckoo contour smoothness is less than a predetermined cuckoo contour smoothness.

[0080] Step S254: If the first azalea contour smoothness is less than the predetermined azalea contour smoothness, a first azalea contour enhancement instruction is generated.

[0081] Step S255: performing cuckoo contour recognition on the first standard remote sensing image sample based on the first cuckoo contour enhancement instruction to determine a first standard contour recognition result.

[0082] Step S256: performing deviation detection on the first cuckoo contour recognition result according to the first standard contour recognition result to determine the first cuckoo contour deviation detection result.

[0083] Step S257: performing contour enhancement processing on the first alpine rhododendron image sample according to the first rhododendron contour deviation detection result.

[0084] Specifically, contour recognition work is carried out based on the first alpine rhododendron image sample that has been obtained. By using edge detection algorithms, such as the Canny algorithm, contour detection is performed on the alpine rhododendron part of the image, and the edge contour of the alpine rhododendron is outlined, thereby obtaining the first rhododendron contour recognition result. This result preliminarily presents the contour shape of the alpine rhododendron in the image.

[0085] The smoothness of the first cuckoo contour is evaluated based on the recognition result of the first cuckoo contour, and then the smoothness of the first cuckoo contour is obtained. First, the recognition result of the first cuckoo contour is presented as a set of contour points, each of which has coordinate information in a two-dimensional plane. Then, starting from these contour points, the vectors between adjacent points are calculated, that is, the coordinates of the latter point are subtracted from the coordinates of the previous point to obtain the direction and distance between the two adjacent points. At the same time, considering the closedness of the contour, the vector between the last point and the first point needs to be calculated. After completing the vector calculation, the angle between adjacent vectors is calculated using the principle of vector dot product. First, the modulus of each vector, that is, the length of the vector, is calculated, and then according to the vector dot product formula, the cosine value of the angle is calculated by multiplying the dot product value and the modulus of the two vectors, and then the actual angle of the angle is obtained by using the inverse cosine function. After the angles of all adjacent vectors are calculated, these angles are averaged to obtain the average angle. Finally, based on the relationship between the average angle and pi, a value between 0 and 1 is obtained by subtracting the ratio of the average angle to pi from 1. The larger the value, the smoother the contour. This value is the smoothness of the first rhododendron contour, which can intuitively reflect the smoothness of the current alpine rhododendron contour and provide a key quantitative basis for subsequent judgment and processing.

[0086] The first rhododendron contour smoothness is compared with a preset rhododendron contour smoothness, which is a reference standard determined based on a large amount of alpine rhododendron image data and actual application needs, and is used to determine whether the smoothness of the current contour meets the requirements.

[0087] If the first rhododendron contour smoothness is less than the predetermined rhododendron contour smoothness, this indicates that the current alpine rhododendron contour is not smooth enough and has many jagged edges or discontinuous parts. At this time, a first rhododendron contour enhancement instruction will be generated to further optimize the contour to make it more in line with the actual situation.

[0088] Based on the generated first rhododendron contour enhancement instruction, the rhododendron contour recognition operation is performed again on the first standard remote sensing image sample. By adjusting the parameters of the recognition algorithm and applying image enhancement technology, the alpine rhododendron contour in the image is re-recognized to obtain more accurate and clear contour information, and finally the first standard contour recognition result is determined.

[0089] Based on the first standard contour recognition result, the first cuckoo contour recognition result is detected for deviation. By comparing the difference between the two contours, calculating the position deviation, shape difference and other indicators of each point on the contour, the first cuckoo contour deviation detection result is determined, so as to understand the degree of deviation and specific deviation between the first cuckoo contour recognition result and the standard contour.

[0090] According to the first rhododendron contour deviation detection result, the first alpine rhododendron image sample is subjected to contour enhancement processing. According to the specific situation of the deviation, the corresponding image restoration and edge smoothing technology is used to optimize and adjust the contour of the alpine rhododendron image to make the contour more accurate and smooth, thereby improving the quality and usability of the alpine rhododendron image.

[0091] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0092] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0093] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. An intelligent identification method of alpine rhododendron based on drone sensors, characterized in that: The method comprises: Obtaining a sample set of UAV remote sensing images of the alpine rhododendron planting area and an alpine rhododendron identification sample set, and performing identification modality classification on the alpine rhododendron identification sample set to obtain Q rhododendron identification modal areas, where Q is a positive integer greater than 1; Inputting the UAV remote sensing image sample set into the alpine rhododendron detection and segmentation module to obtain the alpine rhododendron image sample set; A cuckoo recognition learning channel is introduced to perform loss optimization learning on the alpine cuckoo image sample set and the Q cuckoo recognition modal areas to generate a cuckoo Q modal recognition unit; The GAN network and the predetermined remote sensing scene distribution are introduced to perform generalized reinforcement learning on the cuckoo Q modal recognition unit to obtain a cuckoo modal recognition integrated module, and the alpine cuckoo detection and segmentation module is combined to build an alpine cuckoo recognition module; According to the drone sensor, a remote sensing image of the alpine rhododendron planting area is obtained, and the remote sensing image is input into the alpine rhododendron recognition module to generate a rhododendron Q modal recognition result.

2. The intelligent identification method of alpine rhododendron based on drone sensor according to claim 1 is characterized in that: A cuckoo recognition learning channel is introduced to perform loss optimization learning on the alpine cuckoo image sample set and the Q cuckoo recognition modal areas to generate a cuckoo Q modal recognition unit, including: The cuckoo recognition learning channel includes P cuckoo recognition learners and a cuckoo recognition loss calculator, wherein P is a positive integer greater than 1; Extracting the qth cuckoo identification modal area according to the Q cuckoo identification modal areas, wherein q is a positive integer, 1≤q≤Q; Based on the azalea recognition loss calculator, according to the alpine azalea image sample set and the qth azalea recognition modal area, the P azalea recognition learners are subjected to loss optimization learning to generate the qth modal azalea recognition model; Taking the output data set of the qth modal cuckoo recognition model as input information and the qth cuckoo recognition modal area as output information, training the qth modal cuckoo recognition fusion model; The qth modal cuckoo recognition model is merged with the input layer of the qth modal cuckoo recognition fusion model to generate a cuckoo qth modal recognition unit, and the cuckoo qth modal recognition unit is added to the cuckoo Q modal recognition unit.

3. The intelligent identification method of alpine rhododendron based on drone sensor as claimed in claim 2 is characterized in that: Based on the cuckoo recognition loss calculator, according to the alpine cuckoo image sample set and the qth cuckoo recognition modal area, loss optimization learning is performed on the P cuckoo recognition learners to generate the qth modal cuckoo recognition model, including: According to the alpine rhododendron image sample set and the qth rhododendron recognition modal area, supervised training is performed on the P rhododendron recognition learners respectively, and each time a predetermined number of training times is trained, P rhododendron recognition loss coefficients are calculated according to the rhododendron recognition loss calculator, wherein the rhododendron recognition loss calculator includes a rhododendron recognition loss calculation formula, and the rhododendron recognition loss calculation formula is: ； Among them, LOSS represents the loss coefficient of cuckoo recognition, M represents the scheduled number of training times, m represents the m-th training, M and m are both positive integers, 1≤m≤M, SUOm represents the number of cuckoo recognition prediction samples in the m-th training, and SUXm represents the number of correct samples of cuckoo recognition prediction in the m-th training; If the P cuckoo recognition loss coefficients are less than the cuckoo recognition loss threshold, generating P same-modal cuckoo recognition models; The P same-modal cuckoo recognition models are connected as parallel nodes to generate the q-th modal cuckoo recognition model.

4. The intelligent identification method of alpine rhododendron based on drone sensor as claimed in claim 1 is characterized in that: The GAN network and the predetermined remote sensing scene distribution are introduced to perform generalized reinforcement learning on the cuckoo Q modal recognition unit to obtain a cuckoo modal recognition integrated module. Combined with the alpine cuckoo detection and segmentation module, an alpine cuckoo recognition module is constructed, including: Based on the GAN network, data expansion is performed on the alpine rhododendron image sample set according to the predetermined remote sensing scene distribution ratio to obtain an expanded rhododendron image sample set; Based on the mapping relationship between the alpine rhododendron image sample set and the Q rhododendron recognition modal areas, a rhododendron image-recognition modal mapping network is established; Based on the cuckoo image-recognition modality mapping network, the expanded cuckoo image sample set is migrated and annotated according to the Q cuckoo recognition modality areas to generate an expanded cuckoo recognition modality sample set; Performing generalized reinforcement learning on the cuckoo Q modality recognition unit according to the expanded cuckoo image sample set and the expanded cuckoo recognition modality sample set to generate the cuckoo modality recognition integrated module; The azalea modal recognition integration module and the alpine azalea detection and segmentation module are connected to obtain the alpine azalea recognition module.

5. The intelligent identification method of alpine rhododendron based on drone sensor as claimed in claim 4 is characterized in that: Based on the GAN network, data expansion is performed on the alpine rhododendron image sample set according to the predetermined remote sensing scene distribution ratio to obtain an expanded rhododendron image sample set, including: Collecting remote sensing scene information of the alpine rhododendron image sample set to obtain a remote sensing scene sample set; Calculating a remote sensing scene distribution ratio according to the remote sensing scene sample set; The remote sensing scene distribution optimization target is determined according to the gap between the predetermined remote sensing scene distribution ratio and the remote sensing scene distribution ratio; Based on the remote sensing scene distribution optimization goal, the remote sensing scene sample set is expanded to obtain an expanded remote sensing scene sample set; Based on the expanded remote sensing scene sample set, data expansion is performed on the alpine rhododendron image sample set according to the GAN network to obtain the expanded rhododendron image sample set.

6. The intelligent identification method of alpine rhododendron based on drone sensor according to claim 1 is characterized in that: The UAV remote sensing image sample set is input into the alpine rhododendron detection and segmentation module to obtain the alpine rhododendron image sample set, including: The alpine rhododendron detection and segmentation module includes a pre-processing unit, a rhododendron pixel detection unit and a rhododendron background separation unit; Inputting the UAV remote sensing image sample set into the preprocessing unit to obtain a standard remote sensing image sample set, wherein the preprocessing unit includes a radiation correction model, a geometric correction model and a Gaussian filter; Traversing the standard remote sensing image sample set, extracting a first standard remote sensing image sample; Inputting the first standard remote sensing image sample into the azalea pixel detection unit to obtain the azalea pixel distribution of the first image; Based on the rhododendron pixel distribution of the first image, performing background separation processing on the first standard remote sensing image sample according to the rhododendron background separation unit to obtain a first alpine rhododendron image sample; The first alpine rhododendron image sample is added to the alpine rhododendron image sample set.

7. The intelligent identification method of alpine rhododendron based on drone sensor as claimed in claim 6 is characterized in that: Inputting the first standard remote sensing image sample into the cuckoo pixel detection unit to obtain the cuckoo pixel distribution of the first image includes: The cuckoo pixel detection unit includes a cuckoo pixel prediction model and a cuckoo pixel discrimination model; Obtaining a plurality of pixel samples of the first standard remote sensing image sample; Inputting the plurality of pixel samples into the azalea pixel prediction model to obtain a plurality of azalea pixel prediction coefficients, wherein each azalea pixel prediction coefficient is used to characterize the probability that each pixel sample belongs to an alpine azalea pixel; Inputting the plurality of cuckoo pixel prediction coefficients into the cuckoo pixel discrimination model to obtain a plurality of cuckoo pixel discrimination results, wherein the cuckoo pixel discrimination model includes predetermined cuckoo pixel prediction coefficients; Based on the multiple cuckoo pixel discrimination results, cuckoo pixels are extracted from the first standard remote sensing image sample to generate the cuckoo pixel distribution of the first image.

8. The intelligent identification method of alpine rhododendron based on drone sensor as claimed in claim 6 is characterized in that: Based on the rhododendron pixel distribution of the first image, the first standard remote sensing image sample is subjected to background separation processing by the rhododendron background separation unit to obtain a first alpine rhododendron image sample, further comprising: Performing contour recognition according to the first alpine rhododendron image sample to obtain a first rhododendron contour recognition result; Performing a smoothness evaluation according to the first cuckoo contour recognition result to obtain the first cuckoo contour smoothness; Determining whether the first cuckoo contour smoothness is less than a predetermined cuckoo contour smoothness; If the first cuckoo contour smoothness is less than the predetermined cuckoo contour smoothness, generating a first cuckoo contour enhancement instruction; Performing cuckoo contour recognition on the first standard remote sensing image sample based on the first cuckoo contour enhancement instruction to determine a first standard contour recognition result; Performing deviation detection on the first cuckoo contour recognition result according to the first standard contour recognition result to determine a first cuckoo contour deviation detection result; The first alpine rhododendron image sample is subjected to contour enhancement processing according to the first rhododendron contour deviation detection result.

Images