Plant quadrat investigation method based on deep learning

Abstract

The invention discloses a plant quadrat investigation method based on deep learning, and the method comprises the steps: manually selecting a sampling scheme, carrying out the network connection of a sampling image transmission device and a carrier, and transmitting the sampling scheme; receiving picture data from the sampling picture transmission equipment and the carrier to obtain a sampling data set; analyzing the sampling data set to obtain a rough recognition result set; according to the sampling scheme, carrying out weight addition calculation on the rough recognition result set to obtain a high-precision recognition block, and analyzing the high-precision recognition block to obtain a final quadrat survey result; a species competition database is called according to the final quadrat investigation result, a simulation environment is constructed, and the final quadrat investigation result is introduced and a risk report is obtained; and generating an optional sampling scheme according to the final quadrat survey result and the risk report, and merging the optional sampling scheme into the sampling scheme.

Description

Technical Field

[0001] The present invention relates to the field of quadrat survey, and specifically to a plant quadrat survey method based on deep learning. Background Art

[0002] Plant quadrat surveys are the scientific foundation for biodiversity conservation and ecosystem management. By counting the species, abundance, and spatial distribution of plants within a specific area, they provide critical data for developing conservation strategies. This work plays an irreplaceable role in monitoring the dynamics of endangered species and assessing the effectiveness of ecological restoration.

[0003] Drones and mobile robots have significantly improved the efficiency of plot surveys. Their multispectral sensors can rapidly acquire surface image data over large areas. These devices enable automated trajectory planning and image acquisition in complex terrain, expanding the daily survey area to more than ten times that of traditional manual methods and overcoming accessibility limitations.

[0004] The current mainstream approach uses a "pre-set sampling grid - automated data collection by unmanned equipment - and offline manual analysis" process: The survey area is first divided according to fixed rules, then drones perform standardized imagery. Finally, professionals perform species identification and report generation based on the raw images. This approach establishes a foundational automation framework.

[0005] Existing technologies are unable to proactively identify ecological risks. They require manual, repeated comparisons of historical data to identify threats such as species competition. Single analyses can take weeks and can easily miss early risks. Data silos lead to repeated identification of similar issues in different regions, significantly increasing workload and delaying intervention opportunities, making them unable to meet the dynamic needs of ecological protection. Summary of the Invention

[0006] This application provides a plant sample survey method based on deep learning, which is used to solve the technical problem in the existing technology that the active discovery risk of sample survey requires manual repeated identification of a large amount of data.

[0007] In view of the above problems, this application provides a plant sample survey method based on deep learning.

[0008] The present application provides a plant sample survey method based on deep learning, which includes: manually selecting a sampling plan, connecting a sampling image transmission device and a carrier to a network, and sending the sampling plan; receiving image data from the sampling image transmission device and the carrier to obtain a sampling data set; analyzing the sampling data set to obtain a coarse recognition result set; performing weighted additional calculation on the coarse recognition result set according to the sampling plan to obtain a high-precision recognition block, analyzing the high-precision recognition block to obtain a final sample survey result; retrieving a species competition database according to the final sample survey result, constructing a simulation environment, bringing the final sample survey result into and obtaining a risk report; generating an optional sampling plan according to the final sample survey result and the risk report, and incorporating the optional sampling plan into the sampling plan.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] The embodiments of the present application proactively identify ecological threat relationships by manually presetting target species parameters and weight calculation mechanisms, combining confidence assessment with dynamic threshold decision-making, utilizing species competition databases and risk quantification models; generating intelligent sampling schemes based on multi-dimensional parameters, and relying on human-computer interactive learning mechanisms to continuously optimize scheme recommendation strategies to reduce manual workload.

[0011] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] 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.

[0013] Figure 1 A logical diagram provided for this application. DETAILED DESCRIPTION

[0014] This application provides a plant sample survey method based on deep learning, which is used to solve the technical problem that the existing sample survey requires manual repeated identification of large amounts of data for active discovery risks.

[0015] After introducing the basic principles of the present application, the technical solutions in the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. It should also be noted that, for the sake of ease of description, only the parts related to the present application, not all, are shown in the accompanying drawings.

[0016] Example 1

[0017] This application provides a plant quadrat survey method based on deep learning, which includes:

[0018] S100: Manually select a sampling plan, connect the sampling image transmission device and carrier to the network, and send the sampling plan;

[0019] Image transmission devices and carriers here refer to camera-equipped drones or ground-based mobile robots that can transmit video or images via network transmission. This solution only involves receiving data and sending commands. A sampling plan refers to a preset scheme that generates the move-shoot-transmit-reset command. This command can be implemented using, but is not limited to, third-party control software or the unmanned device's built-in control system.

[0020] For example, in a wetland reserve, ecologists selected the "herbaceous wetland" template and set the grid to 5×5m. The system connected to the DJI M300 drone and sent the first round of sampling coordinates to the airborne terminal to execute the move-shoot-transmit-reset instructions.

[0021] S200: receiving image data from the sampling image transmission device and the carrier to obtain a sampling data set;

[0022] Receive and store the returned image or video data, and extract the video data to obtain the image data. This step can rely on any feasible data transmission device or method. This solution only involves the requirements of data reception and instruction sending without modification, so it will not be described in detail here. All image data is collected into a sample data set.

[0023] Optionally, an existing frame-taking optimization algorithm is used to take frames of the video to avoid obtaining a large number of repeated invalid frames.

[0024] S300: Analyze the sampled data set to obtain a rough recognition result set;

[0025] Normalization is done to facilitate subsequent processing by adjusting the data to a uniform format. This includes, but is not limited to, illumination correction to eliminate brightness differences caused by alternating sunny and cloudy weather. For example, Retinex theory can be used to decompose the illumination / reflection components. Geometric correction eliminates lens distortion, and the normalized data set is output as a standardized image matrix.

[0026] Step S300 in the method provided in the embodiment of the present application includes:

[0027] S310: performing standardization processing on the sampled data set to obtain a normalized data set;

[0028] S320: Extract key features from the normalized data set to obtain a feature set;

[0029] S330: Execute species classification on the feature set through a convolutional network to obtain a rough recognition result;

[0030] S340: All the coarse recognition results are taken as a coarse recognition result set.

[0031] Key features include morphological features and texture features:

[0032] Morphological features, leaf contours are obtained through Canny edge detection. Texture features use the LBP (local binary pattern) operator to calculate the vein distribution entropy (window size 16×16). Multispectral features extract the NDVI vegetation index (ratio of near-infrared and red light bands). The feature set includes quantitative indicators such as contour curvature and texture complexity. For example, for the image of Davidia involucrata leaves, the key features extracted include: contour curvature 0.87 (close to a circle is 1.0), LBP entropy value 2.35 (smooth leaf surface <1.0, complex leaf veins >3.0), and NDVI value 0.81 (healthy vegetation >0.6).

[0033] The convolutional network uses an improved EfficientNet-B4 architecture. The backbone network contains 18 MBConv modules (expansion rate 1.25). Transfer learning loads weights pre-trained on the iNaturalist dataset. The output layer is a 512-category species classifier with softmax activation. The coarse recognition results include the species ID (such as Davidia_involucrata), confidence score (0-1 probability value), and spatial location (based on GPS UTM coordinate system encoding). For example: The Davidia involucrata samples in the block were identified by the network, and the top-3 results output were: Davidia_involucrata (0.89), Nyssa_sinensis (0.07), Camptotheca_acuminata (0.03). The confidence threshold was set to 0.75, so Davidia involucrata was adopted as the valid identification.

[0034] S400: performing weighted additional calculation on the coarse recognition result set according to the sampling scheme to obtain high-precision recognition blocks, analyzing the high-precision recognition blocks to obtain final sample plot survey results;

[0035] Step S400 in the method provided in the embodiment of the present application includes:

[0036] S410: Manually preset target species, priorities, and ecological sensitivity parameters;

[0037] S420: Constructing a weighted additional calculation formula based on the target species, priority, and ecological sensitivity parameters;

[0038] S430: Bring the coarse recognition result set into the weight addition calculation environment to obtain a secondary recognition score, wherein the weight addition calculation formula is as follows:

[0039] W=0.6*Ptarget+0.4*Seco

[0040] Where W is the secondary identification score, Ptarget is the priority preset quantitative value, and Seco is the ecological sensitivity preset quantitative value;

[0041] S440: Substitute the feature set into the confidence calculation formula to calculate the confidence score and obtain the confidence result, wherein the confidence calculation formula is as follows:

[0042] Crough = w1*A contour + w2*A texture (w1=0.7, w2=0.3)

[0043] Where Crough is the confidence score, Acontour is the intersection-over-union ratio of the leaf contour and the standard template, reflecting the morphological matching degree (the closer the value is to 1, the higher the matching degree), Atexture is the cosine similarity of the vein texture (the closer the value is to 1, the more consistent the texture features), w1 and w2 are weight coefficients. These are only used as default setting parameters here, and the coefficients can be manually reallocated and adjusted according to actual preferences;

[0044] S450: Manually setting a weight control threshold and a confidence control threshold;

[0045] S460: If the secondary recognition score is lower than the weight control threshold and the confidence result is higher than the confidence control threshold, the coarse recognition result is still included in the coarse recognition result set; if any one of the conditions of the secondary recognition score being higher than the weight control threshold and the confidence result being lower than the confidence control threshold is met, the coarse recognition result is included in the high-precision recognition block.

[0046] Target species refer to the plant species that users are focused on and are set by ecologists based on the survey objectives.

[0047] Priority is a quantitative value assigned on a scale of 0-10 based on the economic and scientific value of the species. For example, the timber value of Davidia involucrata is 9, which is manually preset based on the survey requirements. Ecological sensitivity is mapped based on the IUCN Red List classification and is scored on a scale of 1-5, with Critically Endangered species being 5. Parameters are entered through an interactive interface or other methods and can be stored in a parameter configuration library or cloud-based network. For example, in a survey, ecologists preset the target species as Davidia involucrata, with a value of 9 for high ornamental value and an ecological sensitivity of 5 for Critically Endangered species.

[0048] The weighting formula assigns 60% weight to economic / scientific value, driving optimal resource allocation. 40% weight considers ecological protection needs, balancing development and conservation. The coefficient adjustment mechanism determines the optimal ratio through regression analysis of historical data.

[0049] The secondary identification score ranks the coarse identification results by strategic importance, triggering a high-precision review. For example, the Davidia involucrata record (ID: J9-012) in the Grid-9 block of the sample plot received a score of W = 7.4 and was marked as a key review object.

[0050] The confidence score is a numerical value used to measure the accuracy of recognition. It is used to infer the validity of the judgment result. Results with insufficient confidence scores cannot be guaranteed to be correct, and manual verification is required.

[0051] S500: Retrieving a species competition database based on the final quadrat survey results, building a simulation environment, and applying the final quadrat survey results to obtain a risk report;

[0052] Step S500 in the method provided in the embodiment of the present application includes:

[0053] S510: traversing all species pairs in the species competition database according to the final quadrat survey results, applying the marking rules in the database, and identifying all risk plants corresponding to the target plant;

[0054] S520: Count the number of risky plants for each target plant and bring it into the risk quantification formula to obtain a risk quantification result. The risk quantification formula is as follows:

[0055]

[0056] Where R represents the risk quantification result, Njr represents the number of risk plants of target plant j, Aj represents the competitive impact coefficient of risk plants on species j, Njt represents the number of target plant j itself, and Sj represents the ecological sensitivity of species j;

[0057] S530: Manually set a risk threshold. If the risk quantification result is greater than the risk threshold, the target plant and the risk plant are added to the risk report; otherwise, if the risk quantification result is less than the risk threshold, the data is generated into a manual report and used to adjust the ecological sensitivity.

[0058] There are numerous competitive relationships between species. The Species Competition Database stores competitive relationships between plants, exemplified by the relationship between Davidia involucrata and Tung oil tree. Data sources include, but are not limited to, ecological literature, historical survey reports, manual data entry, and other cloud data access. For example, during a survey of the Davidia involucrata reserve, a systematic database search revealed a competitive relationship between Tung oil tree and Tung oil tree, with a competition coefficient of 0.7.

[0059] S600: Generate an optional sampling plan based on the final quadrat survey results and the risk report, and incorporate the optional sampling plan into the sampling plan:

[0060] Step S600 in the method provided in the embodiment of the present application includes:

[0061] S610: Arrange all blocks within the sampling range based on risk quantification results, high-precision identification block markers, and ecological sensitivity parameters, and generate multiple survey plans with different arrangement methods;

[0062] S620: Arrange and adjust the plans according to learning preferences;

[0063] S630: Manually select the arranged survey plans to obtain manually retained optional plans;

[0064] S640: Adjust the learning preference strategy according to the manually reserved optional scheme, and mark the manually reserved optional scheme as the optional sampling scheme.

[0065] This step uses algorithms to integrate ecological risk data with historical experience to dynamically adjust sampling plans. Based on risk quantification results, the system marks areas of high competitive pressure, accurately identifies blocks and ecological sensitivity parameters, and uses a multi-objective optimization algorithm to generate diverse plans.

[0066] Learning preference adjustment is the core adaptive mechanism. The system dynamically updates the solution weight coefficient by analyzing the user's historical selection records, such as 80% selection of high-risk priority solutions and execution effect data (such as low review rate solutions are marked as reliable). For example, when the user continuously selects a certain type of solution, its corresponding weight increases from 0.5 to 0.65, making subsequent recommendations more in line with actual needs.

[0067] During the human-machine collaborative decision-making phase, the system provides a visual comparison dashboard showcasing key metrics for each solution (estimated time, risk coverage, and equipment energy consumption), and links historical execution performance data. For example, it displays "The missed detection rate for similar solutions last month was 5.2%." After experts manually select solutions based on ecological protection goals, closed-loop optimization is initiated: selected solutions are tagged and stored, and the strategy is dynamically calibrated based on actual execution results.

[0068] Optionally, if a solution's actual risk coverage exceeds expectations by 5%, the recommended weight of similar solutions automatically increases by 0.08. If a device's energy consumption exceeds expectations by 20%, adjustments to the neighboring block merging algorithm are triggered. This entire process forms a continuous evolutionary cycle of "solution generation → preference learning → manual decision-making → performance feedback," preserving expert judgment while quantifying implicit needs.

[0069] Those skilled in the art will understand that the various numerical numbers such as the first and second involved in this application are only for the convenience of description, and are not used to limit the scope of this application, nor do they indicate the order of precedence. "And / or" describes the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one" refers to one or more. At least two refers to two or more. "At least one", "any one" or similar expressions refer to any combination of these items, including any combination of single items (individuals) or plural items (individuals). For example, at least one item (individual, kind) of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0070] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in this application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions

[0071] The instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0072] The various illustrative logic units and circuits described in this application may be implemented or operated by a design comprising a general purpose processor, a digital signal processor, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof. The general purpose processor may be a microprocessor, which may alternatively be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented by a combination of computing devices, such as a digital signal processor and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a digital signal processor core, or any other similar configuration.

[0073] The steps of the method or algorithm described in this application can be directly embedded in hardware, software units executed by a processor, or a combination of the two. The software units can be stored in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM or any other form of storage medium in the art. Exemplarily, the storage medium can be connected to the processor so that the processor can read information from the storage medium and write information to the storage medium. Optionally, the storage medium can also be integrated into the processor. The processor and the storage medium can be arranged in an ASIC, and the ASIC can be arranged in a terminal. Optionally, the processor and the storage medium can also be arranged in different components in the terminal. These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are performed on the computer or other programmable device to produce computer-implemented processing, so that the instructions executed on the computer or other programmable device provide for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0074] Although the present application has been described with reference to specific features and embodiments thereof, it will be apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and drawings are merely illustrative 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, those skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, the present application is intended to include such modifications and variations as fall within the scope of the present application and its equivalents.

Claims

1. A plant quadrat survey method based on deep learning, characterized in that: The method comprises: Manually select the sampling plan, connect the sampling image transmission equipment and carrier through the network, and send the sampling plan; Receiving image data from the sampling image transmission device and the carrier to obtain a sampling data set; Analyzing the sampled data set to obtain a coarse recognition result set; According to the sampling scheme, weighted additional calculation is performed on the coarse identification result set to obtain high-precision identification blocks, and the high-precision identification blocks are analyzed to obtain final sample plot survey results; Retrieving a species competition database based on the final quadrat survey results, building a simulation environment, and applying the final quadrat survey results to obtain a risk report; An optional sampling plan is generated based on the final sample survey results and the risk report, and the optional sampling plan is incorporated into the sampling plan.

2. A plant quadrat survey method based on deep learning according to claim 1, characterized in that: The analyzing the sampled data set to obtain a rough recognition result set includes: performing standardization processing on the sampled data set to obtain a normalized data set; Extracting key features from the normalized data set to obtain a feature set; Performing species classification on the feature set through a convolutional network to obtain a rough recognition result; All coarse recognition results are taken as a coarse recognition result set.

3. A plant quadrat survey method based on deep learning according to claim 2, characterized in that: The step of performing weighted additional calculation on the coarse recognition result set to obtain a high-precision recognition block includes: Artificially preset target species, priorities, and ecological sensitivity parameters; Construct a weighted additional calculation formula based on the target species, priority and ecological sensitivity parameters; The coarse recognition result set is brought into the weight addition calculation environment to obtain the secondary recognition score, wherein the weight addition calculation formula is as follows: W=0.6*Ptarget+0.4*Seco Where W is the secondary identification score, Ptarget is the priority preset quantitative value, and Seco is the ecological sensitivity preset quantitative value; The feature set is brought into the confidence calculation formula to calculate the confidence score and obtain the confidence result, wherein the confidence calculation formula is as follows: Crough = w1*A contour + w2*A texture (w1=0.7, w2=0.3) Where Crough is the confidence score, A contour is the intersection-over-union ratio of the leaf contour and the standard template, reflecting the morphological matching degree (the closer the value is to 1, the higher the matching degree), A texture is the cosine similarity of the vein texture (the closer the value is to 1, the more consistent the texture features are), and w1 and w2 are weight coefficients; Manually set weight control threshold and confidence control threshold; If the secondary recognition score is lower than the weight control threshold and the confidence result is higher than the confidence control threshold, the coarse recognition result is still included in the coarse recognition result set; if any one of the conditions of the secondary recognition score being higher than the weight control threshold and the confidence result being lower than the confidence control threshold is met, the coarse recognition result is included in the high-precision recognition block.

4. A plant sample survey method based on deep learning according to claim 3, characterized in that: The method of retrieving a species competition database based on the final quadrat survey results, and bringing the final quadrat survey results into and obtaining a risk report includes: Traversing all species pairs in the species competition database based on the final sample survey results, applying the marking rules in the database, and identifying all risk plants corresponding to the target plant; The number of risky plants for each target plant is counted and brought into the risk quantification formula to obtain the risk quantification result. The risk quantification formula is as follows: Where R represents the risk quantification result, Njr represents the number of risk plants of target plant j, Aj represents the competitive impact coefficient of risk plants on species j, Njt represents the number of target plant j itself, and Sj represents the ecological sensitivity of species j; A risk threshold is manually set. If the risk quantification result is greater than the risk threshold, the target plant and the risk plant are added to the risk report; otherwise, if the risk quantification result is less than the risk threshold, the data is generated into a manual report and used to adjust the ecological sensitivity.

5. The plant sample survey method based on deep learning according to claim 1, characterized in that: Generating an optional sampling plan based on the final quadrat survey results and the risk report, and incorporating the optional sampling plan into the sampling plan, includes: Arrange all blocks within the sampling range according to risk quantification results, high-precision identification block markers, and ecological sensitivity parameters, and generate multiple survey plans with different arrangement methods; Arrange and adjust the programs according to learning preferences; Manually select the arranged survey plans to obtain manually retained optional plans; The learning preference strategy is adjusted according to the manually reserved alternatives, and the manually reserved alternatives are marked as the optional sampling alternatives.

Images