A method and system for monitoring illicit poppy cultivation in a yard

Abstract

The application provides a kind of illegal poppy monitoring method and system in courtyard, belongs to intelligent monitoring technical field, this method includes: unmanned aerial vehicle shoots target courtyard remote sensing image, obtains splicing vector with scene weight by single dimension reconstruction, feature extraction, disassembles single dimension vector and calculates contrast coefficient group to judge poppy matching element, respectively adopts general poppy identification model, courtyard exclusive poppy growth cycle identification model according to label to carry out image annotation, finally, redundant processing is done to annotated image and feature weighted fusion is obtained to obtain monitoring result containing planting position, area, growth cycle. Significantly improve the accuracy and efficiency of illegal poppy monitoring in courtyard.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, and in particular to a method and system for monitoring the illegal planting of poppies in courtyards. Background Technology

[0002] Drone remote sensing combined with image recognition technology has been widely used in the monitoring of illegal poppy cultivation. However, existing technologies are mostly designed for monitoring large-scale poppy cultivation in the wild, and have many technical shortcomings for monitoring small-scale, obstructed illegal poppy cultivation in residential courtyards. Current monitoring methods are simply combinations of conventional functional modules. For example, drone footage is not adjusted for environmental characteristics such as the spatial scale of the courtyard, building obstructions, and mixed vegetation. Image integration and fusion do not consider the problem of incomplete poppy outlines caused by vegetation obstruction, and only perform conventional pixel stitching. The integrated image cannot fully reproduce the growth characteristics of poppies in the courtyard, leading to a higher rate of missed detection of easily obstructed poppy seedlings. Therefore, these methods cannot meet the needs for accurate monitoring of small-scale, obstructed poppy cultivation in courtyards.

[0003] Therefore, this invention proposes a method and system for monitoring the illegal planting of poppies in courtyards. Summary of the Invention

[0004] This invention provides a method and system for monitoring the illegal planting of poppies in courtyards, in order to solve the aforementioned technical problems.

[0005] This invention provides a method for monitoring the illegal cultivation of poppies in a courtyard, comprising: Step 1: Control the drone to capture several remote sensing images of the target courtyard and perform several single-dimensional reconstruction processes on each remote sensing image to obtain a one-to-one corresponding single-dimensional image. At the same time, extract environmental features and existing object features from each remote sensing image and compare it with the standard environment for planting poppies in the courtyard scene and the standard growth of different growth cycles to obtain the stitched vector of the corresponding remote sensing image with courtyard scene feature weights. Among them, there is at least one remote sensing image under different set shooting positions. Step 2: Match the image decomposition method from the type-image analysis lookup table according to the dimension type of the single dimension, and decompose the corresponding single-dimensional image according to the image decomposition method to obtain the single-dimensional vector for monitoring poppies in the courtyard; Step 3: Determine the courtyard feature comparison coefficient group of each single-dimensional vector under the same remote sensing image and the stitched vector respectively, and determine whether there is a poppy matching element in the corresponding single-dimensional vector. If there is, construct the combination sequence of the corresponding single-dimensional vector according to the courtyard spatial feature rules and assign a comprehensive analysis label. If there is no poppy matching element, assign an independent analysis label to the corresponding single-dimensional vector. Step 4: Input the single-dimensional image of the single-dimensional vector with independent analysis labels into the general poppy image recognition model, and output the first image with labels. At the same time, integrate all the combination sequences with comprehensive analysis labels according to the mapping relationship of dimension type to obtain the second image. Then, input the second image into the poppy recognition model with different growth stages of poppies in the courtyard scene in turn, and output the second image with labels in turn. Step 5: Perform redundancy processing on all labeled images from the same shooting location, and then perform feature weighted fusion processing on the redundant images to obtain the monitoring results of illegal poppy planting in the target courtyard, including planting location, area, and growth cycle.

[0006] Preferably, before using remote sensing images of the target courtyard captured by a drone, the process includes: The drones are controlled to take pictures of the courtyard-style planting verification area, which is covered by walls and green plants, in turn at a rotation azimuth angle of 30°. The images to be analyzed are obtained at each rotation azimuth angle. At the same time, the standard images of the courtyard-style planting verification area taken by the normal camera at each rotation azimuth angle are obtained. The image to be analyzed at the same rotation azimuth angle is compared with the standard image at the same location point to obtain the first restored image by proportionally restoring the corresponding image to be analyzed. The courtyard remote sensing-specific abnormal parameter group of each first restored image is obtained and the first abnormal parameter matrix is ​​constructed. The shooting amplitude attitude and courtyard environment interference conditions of the UAV at each rotation azimuth angle are retrieved and matched with the attitude-condition-correction lookup table to obtain correction conditions. Each abnormal parameter in the courtyard remote sensing shooting exclusive abnormal parameter group of the first restored image at the corresponding rotation azimuth angle is fine-tuned. Then, a second abnormal parameter matrix is ​​constructed based on the combination of all fine-tuned parameters. Obtain the first and second anomalous feature vectors of the first and second anomalous parameter matrices, and determine the non-persistent anomalous parameters of the courtyard remote sensing images. Extract the first column vector and the second column vector of each non-stubborn anomaly parameter from the first anomaly parameter matrix and the second anomaly parameter matrix respectively, and calculate the improvement coefficient of the corresponding non-stubborn anomaly parameter based on the first column vector and the second column vector of the same non-stubborn anomaly parameter to obtain an improved sequence group including the non-stubborn anomaly parameter. The improved sequence set is compared with the non-persistent parameters in the standard parameters of the normal camera to obtain the difference sequence set; When the confidence level of the difference sequence group is greater than a preset level, the original camera installed on the drone is replaced by matching the camera to be replaced from the difference-camera lookup table that matches the difference sequence group. Otherwise, it is determined that the original camera installed on the drone does not need to be replaced.

[0007] Preferably, the drone is controlled to capture several remote sensing images of the target courtyard, including: When the drone is shooting at the set shooting position, the drone's maximum amplitude and self-position are captured in real time, and the offset is compared with the current maximum amplitude and current position caused by the current shooting environment. If the contrast offset of the drone remains within the standard constraint range at all times during the shooting process, then the drone is controlled to shoot at the set shooting position according to the set number of shots. Otherwise, capture the instantaneous moments when the comparison offset is not within the standard constraint range, and determine the updated number of shots based on the time interval between the instantaneous moment and the start shooting moment, the extreme brewing variable based on the instantaneous moment and the current shooting accuracy of the drone. The extreme brewing variable is the sum of the drone amplitude change rate caused by the airflow disturbance of the courtyard building and the shooting field of view change rate caused by the occlusion of the courtyard green plants. The drone is controlled to continue taking pictures at the set shooting position according to the updated number of pictures.

[0008] Preferably, a combination sequence of corresponding one-dimensional vectors is constructed according to the spatial feature rules of the courtyard, including: When the poppy matching element is an environment type, obtain the radiation space range of the matching element in the small planting areas such as flower pots, corners, and gaps between green plants in the target courtyard; When the matching element is an object type, the small-area object outline and object features of the matching element in the target courtyard are obtained, and the combination sequence is obtained by sequentially arranging the radiation space range and / or object outline and object features of the matching elements involved in the corresponding one-dimensional vector according to the courtyard space feature rules.

[0009] Preferably, the courtyard feature comparison coefficient set for each single-dimensional vector in the same remote sensing image and the stitched vector is determined separately, including: The single-dimensional vector and the spliced ​​vector are normalized and aligned to reflect the dimensions of poppy monitoring-related features in the courtyard. The missing poppy monitoring-related feature dimensions in the single-dimensional vector are padded with zero values ​​to obtain the aligned single-dimensional vector and the aligned spliced ​​vector. The feature dimensions corresponding to the standard environment and standard growth of poppies in different growth cycles under the courtyard scene in the spliced ​​vector are extracted as core dimensions. The core dimensions are assigned a first weight, and the other non-core dimensions are assigned a second weight, with the first weight being greater than the second weight. Calculate the cosine similarity coefficient and Euclidean distance deviation coefficient of the aligned single-dimensional vector and the aligned concatenated vector in each dimension respectively. Combine the weights of the corresponding dimensions to perform a weighted fusion operation on the two coefficients to obtain the fusion coefficients of each dimension. The fusion coefficients of all dimensions are combined according to the order of the dimension types of the yard poppy monitoring to obtain the control coefficient group corresponding to each single-dimensional vector and the spliced ​​vector. Each fusion coefficient in the control coefficient group is associated with the dimension type and weight assignment information of the corresponding yard poppy monitoring.

[0010] Preferably, the second image is obtained by integrating all combined sequences with comprehensive analysis labels according to the mapping relationship of dimensional types, including: By classifying the combined sequences with comprehensive analysis labels under the same remote sensing image, we can obtain several categories of combined elements; The dimension type with the highest mapping confidence is extracted from the mapping relationship of the corresponding dimension type of the comprehensive analysis label and regarded as the first dimension. The element features based on the first dimension are then used to construct the first contour baseline. Meanwhile, the dimension type that is closest to the feature contour is extracted from the dimension type assigned to the comprehensive analysis label and regarded as the second dimension. At this time, the second contour baseline is obtained based on the element features of the second dimension. The first contour baseline and the second contour baseline are aligned. When the first contour baseline and the second contour baseline are completely consistent, the second contour baseline is used as the initial contour baseline. Otherwise, lock the inconsistency point of the first contour baseline. When the inconsistency point is located inside the second contour baseline, determine the first nearest perpendicular point between the inconsistency point and the second contour baseline, and lock the center point of the straight line connecting the first nearest perpendicular point and the inconsistency point. At the same time, obtain the two second adjacent points through which the tangent of the inconsistency point passes through the second contour baseline, and the curvature direction of the inconsistency point in the second contour baseline. Based on the ratio of the first length between the two adjacent points to the baseline length of the first contour baseline located inside the second contour baseline, and in conjunction with the curvature direction and the angle between the curvature direction and the connecting line, the center point is adjusted along the connecting line. When the inconsistency point is located outside the second contour baseline, the average of the nearest distance difference between all inconsistency points located outside the second contour baseline and the second contour baseline is obtained. At the same time, the baseline length between the corresponding inconsistency point and the second contour baseline and the nearest coincident point on the second contour baseline is obtained and regarded as the first length. The inconsistency points are adjusted a second time along a line perpendicular to the second profile baseline based on the average value, the shortest distance length, and the first length. A preliminary contour baseline is obtained by integrating the adjusted points and the unadjusted points; For each type of remaining combination element, supplementary features for monitoring poppies in the courtyard are extracted according to the corresponding dimension type. In addition, the corresponding supplementary features are preferentially superimposed with the poppy feature parameters of the courtyard shading area for radiation processing, which is applied within the framework of the preliminary contour baseline. When all remaining combination elements have been radiated, the second image is obtained.

[0011] Preferably, the radiative processing of the corresponding supplementary features is applied within the framework of the initial contour baseline, including: The mapping similarity between the remaining combined elements of the courtyard poppy monitoring dimension type and the first and second dimensions is calculated based on the cosine similarity algorithm, and then matched with the level comparison table to obtain the mapping level. When the mapping level is high priority radiation, the leaf contour and plant height parameters that are strongly correlated with the monitoring of poppies in the courtyard are weighted and superimposed onto the core area of ​​the preliminary contour baseline. When the mapping level is medium priority radiation, the corresponding supplementary features are scaled and normalized, and pixel-level weighted fusion is performed to inject the secondary core region of the preliminary contour baseline. When the mapping level is low-priority radiation, the corresponding supplementary features are used as background enhancement information and low-intensity superimposed on the edge transition region of the initial contour baseline.

[0012] This invention provides a monitoring system for illegal poppy cultivation in courtyards, comprising: The reconstruction and stitching module is used to control the drone to capture several remote sensing images of the target courtyard and perform several single-dimensional reconstruction processes on each remote sensing image to obtain a one-to-one corresponding single-dimensional image. At the same time, environmental features and existing object features are extracted from each remote sensing image, and compared with the standard environment for planting poppies in the courtyard scene and the standard growth of different growth cycles to obtain the stitching vector of the corresponding remote sensing image with courtyard scene feature weights. Among them, there is at least one remote sensing image under different set shooting positions. The single-dimensional decomposition module is used to match the image decomposition method from the type-image analysis lookup table according to the dimension type of the single dimension, and to decompose the corresponding single-dimensional image according to the image decomposition method to obtain the single-dimensional vector for monitoring the poppy in the courtyard. The element judgment module is used to determine the courtyard feature comparison coefficient group of each single-dimensional vector under the same remote sensing image and the stitched vector, and to determine whether there is a poppy matching element in the corresponding single-dimensional vector. If there is, the combination sequence of the corresponding single-dimensional vector is constructed according to the courtyard spatial feature rules and a comprehensive analysis label is assigned. If there is no poppy matching element, an independent analysis label is assigned to the corresponding single-dimensional vector. The image annotation module is used to input a single-dimensional image of a single-dimensional vector with an independent analysis label into a general poppy image recognition model and output a first image with annotations. At the same time, all combination sequences with comprehensive analysis labels are integrated in detail according to the mapping relationship of dimension type to obtain a second image. The second image is then input into a poppy recognition model for different growth stages of poppies in a courtyard scene and outputs a second image with annotations in sequence. The monitoring integration module is used to perform redundant processing on all labeled images from the same shooting location, and to perform feature weighted fusion processing on the redundant images of the courtyard scene to obtain monitoring results of illegal poppy planting in the target courtyard, including planting location, area, and growth cycle.

[0013] Compared with the prior art, the beneficial effects of this application are as follows: By setting drone shooting parameters specific to courtyards, clear images of small, obscured planting areas in courtyards were ensured. Feature vectors were constructed by combining courtyard scene feature weights and a dedicated set of control coefficients were designed, improving the accuracy of poppy feature matching. A dual-model recognition system, consisting of a general model for initial screening and a courtyard-specific growth cycle model for fine screening, reduced the false negative and false positive rates in courtyard poppy monitoring. Finally, through image redundancy processing and feature weighted fusion specific to courtyard scenes, the comprehensiveness and accuracy of monitoring results were achieved, significantly improving the accuracy and efficiency of monitoring small-area, obscured illegal poppy cultivation in courtyards.

[0014] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a method for monitoring the illegal planting of poppies in a courtyard, as described in an embodiment of the present invention. Figure 2 This is a structural diagram of a monitoring system for illegal poppy cultivation in a courtyard, as described in an embodiment of the present invention. Detailed Implementation

[0017] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0018] This invention provides a method for monitoring the illegal cultivation of poppies in courtyards, such as... Figure 1 As shown, it includes: Step 1: Control the drone to capture several remote sensing images of the target courtyard and perform several single-dimensional reconstruction processes on each remote sensing image to obtain a one-to-one corresponding single-dimensional image. At the same time, extract environmental features and existing object features from each remote sensing image and compare it with the standard environment for planting poppies in the courtyard scene and the standard growth of different growth cycles to obtain the stitched vector of the corresponding remote sensing image with courtyard scene feature weights. Among them, there is at least one remote sensing image under different set shooting positions. Step 2: Match the image decomposition method from the type-image analysis lookup table according to the dimension type of the single dimension, and decompose the corresponding single-dimensional image according to the image decomposition method to obtain the single-dimensional vector for monitoring poppies in the courtyard; Step 3: Determine the courtyard feature comparison coefficient group of each single-dimensional vector under the same remote sensing image and the stitched vector respectively, and determine whether there is a poppy matching element in the corresponding single-dimensional vector. If there is, construct the combination sequence of the corresponding single-dimensional vector according to the courtyard spatial feature rules and assign a comprehensive analysis label. If there is no poppy matching element, assign an independent analysis label to the corresponding single-dimensional vector. Step 4: Input the single-dimensional image of the single-dimensional vector with independent analysis labels into the general poppy image recognition model, and output the first image with labels. At the same time, integrate all the combination sequences with comprehensive analysis labels according to the mapping relationship of dimension type to obtain the second image. Then, input the second image into the poppy recognition model with different growth stages of poppies in the courtyard scene in turn, and output the second image with labels in turn. Step 5: Perform redundancy processing on all labeled images from the same shooting location, and then perform feature weighted fusion processing on the redundant images to obtain the monitoring results of illegal poppy planting in the target courtyard, including planting location, area, and growth cycle.

[0019] In this embodiment, a drone aerial photography altitude range is set for the courtyard scene. This altitude range is adapted to the typical spatial scale of the courtyard, enabling clear capture of detailed features of small planting areas such as flower pots and corners, while avoiding the problems of insufficient shooting pixels due to excessively high altitudes and limited field of view due to excessively low altitudes, such as for areas ≤ For small courtyard houses, a shooting height of 5m is preferred, targeting areas of 50-200 square meters. For large courtyard houses, an 8m shooting height is preferred, targeting areas of 200-300 square meters. For the courtyard, a shooting height of 10m is preferred.

[0020] The drone's flight control system imports the geographic outline and latitude and longitude coordinates of the target courtyard, automatically plans the latitude and longitude coordinates of 3-5 shooting positions, and the drone autonomously moves to each shooting position in sequence according to the planned coordinates. After reaching each shooting position, it completes hovering and positioning shooting. For example, for a small square courtyard, 3 shooting positions are set: directly above the courtyard, above the northeast corner of the courtyard, and above the southwest corner of the courtyard. For a large courtyard with an irregular outline, 5 shooting positions are set, evenly distributed in the upper area around the courtyard.

[0021] The remote sensing image is a courtyard scene image captured by a high-definition aerial camera mounted on a drone at a set shooting position. The image contains all visual information of the courtyard, including buildings, green plants, planting areas, etc.

[0022] Feature dimensions include color dimension, contour dimension, texture dimension, pixel brightness dimension, etc. For example, color dimension reconstruction processing of a panoramic remote sensing image of a courtyard results in a single-dimensional image that retains only the color feature information in the image. Contour dimension reconstruction processing of the same remote sensing image results in a single-dimensional image that retains only the contour feature information of objects in the image.

[0023] Environmental features are environmental features related to poppy cultivation monitoring extracted from courtyard scenes. Among them, occlusion features include wall occlusion, green plant occlusion, building corner occlusion, etc. Green plant mixture features include the distribution, type, and growth status of green plants such as roses, pothos, and holly in the courtyard. Small planting area features include the location, area, and shape of areas where small-scale planting can be carried out, such as flower pots, flower boxes, courtyard corners, and gaps between green plants.

[0024] Object characteristics are visual features such as shape, color, texture, and plant type extracted from small planting areas in the courtyard. These are the basic characteristics for determining whether a plant is a poppy. For example, the object characteristics extracted from the flowerpot planting area in the courtyard are that the plant is about 15cm tall, the leaves are serrated, the leaves are dark green, and the plant is rosette-shaped.

[0025] The standard environment for planting poppies in a courtyard setting is based on pre-collected environmental characteristics suitable for poppy growth, targeting small-area, shaded planting scenarios in courtyards. This standard includes the types of areas in the courtyard where poppies are easily planted, the surrounding greenery, and the shading conditions. For example, the standard environment for planting poppies in a courtyard is: small planting areas such as flowerpots or gaps between green plants, surrounded by broad-leaved greenery for shading, with semi-shaded light, and a planting area ≤ [missing information]. .

[0026] The standard growth characteristics for different growth stages are the standard morphological characteristics of poppies at different growth stages such as seedling stage, mature stage, and flowering stage. They are stored in advance. For example, the standard growth characteristics for poppy seedlings are: plant height 5-20cm, serrated leaves, rosette-shaped plant, and no stem; the standard growth characteristics for mature plants are: plant height 30-60cm, with a distinct stem, alternate leaves, and lanceolate shape.

[0027] The correlation between various features in the courtyard scene and poppy monitoring was determined by the analytic hierarchy process. The weight values ​​based on the correlation were pre-stored for easy retrieval in subsequent use. Among them, the weight of strongly correlated features such as poppy leaf outline, plant shape, and planting area type was 0.8, while the weight of weakly correlated features such as courtyard wall color and ground material was 0.2.

[0028] The splicing vector is generated by calculating the similarity between the extracted actual environmental features and object features of the courtyard and the standard environment and standard growth features, and then weighting them according to the feature weights of the courtyard scene. The weighted results of each dimension are arranged in a preset feature order to generate a multi-dimensional splicing vector. For example, the splicing vector is: [0.85, 0.21, 0.78, 0.15], where each value represents the weighted comparison result of leaf outline features, wall color features, plant shape features, and ground material features.

[0029] In this embodiment, the type-image analysis lookup table is a pre-constructed lookup table that corresponds one-to-one between dimension types and image decomposition methods, as shown in Table 1: Table 1 Type-Image Analysis Comparison Table

[0030] A one-dimensional vector is a numerical vector formed by decomposing a one-dimensional image according to a matching image decomposition method and arranging the extracted quantized feature values ​​in a preset order. For example, after decomposing a color-dimensional one-dimensional image by RGB channels, the pixel mean, variance and other feature values ​​of each channel are extracted and arranged to form a one-dimensional vector of [125,36,89,21,15,9].

[0031] The poppy monitoring feature dimension is a set of feature dimensions that are strongly correlated with poppy identification and are designed for monitoring poppies in courtyards. This set excludes feature dimensions that are irrelevant to poppy monitoring in courtyards, thereby improving the efficiency of vector comparison. For example, the poppy monitoring feature dimensions for courtyard scenes include leaf outline dimension, plant shape dimension, color dimension, and planting area dimension, while excluding irrelevant dimensions such as wall material dimension and ground color dimension.

[0032] The courtyard feature comparison coefficient set is a set of coefficients formed by calculating the fusion coefficients of the single-dimensional vector and the spliced ​​vector in each dimension within the poppy monitoring feature dimension of the courtyard scene, and then arranging them in dimensional order. This coefficient set reflects the feature matching degree between the single-dimensional vector and the spliced ​​vector. For example, the courtyard feature comparison coefficient set is [0.92, 0.87, 0.13], which represent the fusion coefficients of the leaf outline, plant shape, and color dimensions, respectively.

[0033] Poppy matching elements are feature information extracted from a single-dimensional vector that matches the standard features of poppies in a courtyard scene. The criteria for determining matching elements is whether the fusion coefficient in the courtyard feature comparison coefficient group is higher than a preset threshold of 0.8. For example, if the fusion coefficient between the leaf contour feature in the single-dimensional vector and the standard leaf contour feature of a poppy seedling is 0.95, which is higher than the preset threshold, then the leaf contour feature is the poppy matching element.

[0034] In this embodiment, the courtyard spatial feature rules serve as the standard for constructing and arranging combination sequences in courtyard poppy monitoring. Based on the target courtyard's planar coordinate system, planting area type, and poppy growth cycle, the rules are quantified and arranged. This is applicable to constructing combination sequences of poppy matching elements for environmental / object types. The specific execution standards are as follows: Spatial coordinate sorting benchmark: with the northwest corner of the target courtyard as the origin of the plane coordinate (0,0), the positive direction of the X-axis is eastward and the positive direction of the Y-axis is southward, and the coordinates are arranged in order of the center coordinates of the planting area from small to large X-value and from large to small Y-value; Planting area type priority: flower pot planting area (priority 1) > green plant gap planting area (priority 2) > courtyard corner planting area (priority 3) > flower box planting area (priority 4), and within the same coordinate range, they are arranged from high to low priority; Poppy growth cycle priority: seedling stage (priority A) > mature plant stage (priority B) > flowering stage (priority C), and within the same planting area, they are arranged from high to low priority according to growth cycle. Combination sequence dimensional arrangement: The combination sequence of the same poppy matching element is arranged in a fixed dimensional order of outline dimension → plant shape dimension → color dimension → texture dimension → pixel brightness dimension to ensure the consistency of feature integration.

[0035] In this embodiment, the combined sequence is a vector sequence formed by arranging multiple single-dimensional vectors containing poppy matching elements according to the spatial feature rules of the courtyard. This sequence integrates poppy matching feature information of multiple dimensions, such as arranging single-dimensional vectors of color dimension, outline dimension, and plant shape dimension according to the spatial order of the courtyard planting area to form a combined sequence of: [[125,36,89],[0.92,0.87,0.13],[15,21,36]].

[0036] The comprehensive analysis label is a label assigned to a single-dimensional vector and combination sequence containing poppy matching elements. This label is used to distinguish the path of image analysis. Feature data with this label will enter the courtyard-specific growth cycle recognition model for fine screening and analysis. For example, a combination sequence can be labeled with the comprehensive analysis label "comprehensive analysis - flower pot planting area" to clarify the analysis type and the corresponding planting area.

[0037] Independent analysis labels are annotation labels assigned to single-dimensional vectors that do not contain poppy matching elements. These labels are used to distinguish the path of image analysis. Feature data with this label will enter the general poppy image recognition model for initial screening analysis. For example, a single-dimensional vector of a certain color dimension is labeled with the independent analysis label "independent analysis - no matching element".

[0038] The universal poppy image recognition model is a deep learning model for poppy recognition trained on image data from common poppy cultivation scenarios. This model can identify the common morphological features of poppies and is suitable for preliminary screening of poppies. The universal poppy image recognition model trained on the YOLOv8 algorithm can identify poppy plants in images and mark their locations. The image data consists of several poppy cultivation images, including seedling stage (3000 images), mature stage (4000 images), and flowering stage (3000 images). The image resolution is 2K-4K, covering different lighting conditions such as sunny days, cloudy days, and morning and evening.

[0039] The first image is a single-dimensional image with an independent analysis label input into a general poppy image recognition model. The output image is a poppy recognition label. If the model does not recognize a poppy, it is labeled "no poppy". If it recognizes a poppy, it is labeled with information such as the location and shape of the poppy.

[0040] The mapping relationship of dimensional types is a pre-constructed feature association relationship between various dimensional types. This relationship is specific to the monitoring of poppies in the yard and is used to integrate the details of the combined sequence to ensure that the integrated image can completely restore the feature information of poppies in the yard. For example, the mapping relationship between the contour dimension and the color dimension is: the contour point coordinates correspond to the pixel values ​​of the RGB color channels.

[0041] The second image is obtained by integrating the features of the combined sequence with comprehensive analysis labels according to the mapping relationship of the dimension type and reconstructing the pixels. This image integrates poppy matching feature information of multiple dimensions, which is closer to the actual planting scene in the courtyard. For example, by integrating the features of the combined sequence of contour dimension, color dimension and plant shape dimension, the reconstructed image clearly shows the contour and color of poppy seedlings in the flower pot in the courtyard.

[0042] The poppy identification model for different growth stages in courtyard settings is a deep learning model trained using images of poppies at different growth stages in courtyards as training samples. This model is specifically designed for small-area, shaded planting scenarios in courtyards. It consists of three sub-models: seedling stage, mature plant stage, and flowering stage, each adapted to the feature recognition of different growth stages of poppies. This model is suitable for refined screening of poppies. The courtyard poppy seedling identification model, trained based on the YOLOv8 algorithm, can accurately identify poppy seedlings in shaded environments. The training samples consist of several images of poppy planting in courtyards, all with a resolution of 4K-8K, covering planting areas such as flowerpots, corners, and gaps between green plants, as well as environments such as walls and green plant shading. Each image is labeled with information such as poppy location, plant height, number of leaves, and type of planting area.

[0043] Initial screening using the general model: A single-dimensional image with an independent analysis label is input into the general model. If the model outputs a recognition confidence score ≥ 0.7, it is determined to be a suspected poppy area, and the feature information of the image is synchronized to the courtyard-specific model; if the confidence score < 0.7, it is determined to be poppy-free, and is directly labeled "poppy-free". Dedicated model screening: Image features of suspected poppy areas and a second image with comprehensive analysis labels are sequentially input into the garden-specific seedling stage, mature stage, and flowering stage sub-models. The sub-models determine the poppy growth cycle from high to low confidence. If the highest confidence is ≥0.85, it is determined to be poppy in the corresponding growth cycle; if the confidence of all sub-models is <0.85, it is determined to be non-poppy.

[0044] The labeled second image is the image output by the poppy growth cycle recognition model specifically for the courtyard scene after the second image is input sequentially. The model will label the poppy growth cycle, planting location, number of plants and other information based on the recognition results. For example, if a second image is input into the seedling stage sub-model and is identified as a poppy seedling, the output labeled second image will be labeled with the text "Poppy seedling - Flower pot planting area - 3 plants" and the location box information.

[0045] Redundancy processing is a method of removing redundant images with duplicate content and identical annotation information from multiple labeled images taken at the same location. For example, if three labeled images are taken at the same location and the annotation information is "poppy seedlings - flower pot planting area", only one clear image is kept and the other two redundant images are deleted.

[0046] The feature-weighted fusion processing of courtyard scenes is a processing method that integrates the features and annotation information of multiple images by combining the feature weights of the courtyard scene with the image fusion algorithm. The purpose of fusion is to integrate the effective information of each image to obtain complete and accurate feature information of poppy cultivation in the courtyard. For example, feature-weighted fusion is performed on the labeled images of the contour dimension and color dimension at the same shooting location to obtain the complete contour and color features of poppy cultivation, while also integrating the annotation information.

[0047] The monitoring results are comprehensive information on the illegal poppy cultivation in the target courtyard, obtained through image fusion processing. This information includes key details such as the planting location, area, growth cycle, and number of poppies, presented in both text reports and visual images. For example, a text report might read: "Poppy seedlings were found in the flowerpot planting area in the center of the target courtyard, with a planting area of..." There were 5 poppy plants; mature poppy plants were found in the southwest corner of the planting area, covering an area of ​​[missing information]. The package contains two poppy plants and includes a panoramic view of the courtyard with the location of the poppy planted marked.

[0048] The beneficial effects of the above technical solution are as follows: By setting drone shooting parameters specific to courtyards, clear image acquisition of small-area, obscured planting areas in courtyards is ensured; by constructing feature vectors based on courtyard scene feature weights and designing a specific set of control coefficients, the accuracy of poppy feature matching is improved; through a dual-model recognition system of initial screening using a general model and fine screening using a courtyard-specific growth cycle model, the false negative and false positive rates of courtyard poppy monitoring are reduced; and finally, through image redundancy processing and feature weighted fusion specific to courtyard scenes, the comprehensiveness and accuracy of monitoring results are achieved, significantly improving the accuracy and efficiency of monitoring small-area, obscured illegal poppy cultivation in courtyards.

[0049] This invention provides a method for monitoring the illegal planting of poppies in courtyards, which includes the following steps before using a drone to capture remote sensing images of the target courtyard: The drones are controlled to take pictures of the courtyard-style planting verification area, which is covered by walls and green plants, in turn at a rotation azimuth angle of 30°. The images to be analyzed are obtained at each rotation azimuth angle. At the same time, the standard images of the courtyard-style planting verification area taken by the normal camera at each rotation azimuth angle are obtained. The image to be analyzed at the same rotation azimuth angle is compared with the standard image at the same location point to obtain the first restored image by proportionally restoring the corresponding image to be analyzed. The courtyard remote sensing-specific abnormal parameter group of each first restored image is obtained and the first abnormal parameter matrix is ​​constructed. The shooting amplitude attitude and courtyard environment interference conditions of the UAV at each rotation azimuth angle are retrieved and matched with the attitude-condition-correction lookup table to obtain correction conditions. Each abnormal parameter in the courtyard remote sensing shooting exclusive abnormal parameter group of the first restored image at the corresponding rotation azimuth angle is fine-tuned. Then, a second abnormal parameter matrix is ​​constructed based on the combination of all fine-tuned parameters. Obtain the first and second anomalous feature vectors of the first and second anomalous parameter matrices, and determine the non-persistent anomalous parameters of the courtyard remote sensing images. Extract the first column vector and the second column vector of each non-stubborn anomaly parameter from the first anomaly parameter matrix and the second anomaly parameter matrix respectively, and calculate the improvement coefficient of the corresponding non-stubborn anomaly parameter based on the first column vector and the second column vector of the same non-stubborn anomaly parameter to obtain an improved sequence group including the non-stubborn anomaly parameter. The improved sequence set is compared with the non-persistent parameters in the standard parameters of the normal camera to obtain the difference sequence set; When the confidence level of the difference sequence group is greater than a preset level, the original camera installed on the drone is replaced by matching the camera to be replaced from the difference-camera lookup table that matches the difference sequence group. Otherwise, it is determined that the original camera installed on the drone does not need to be replaced.

[0050] In this embodiment, the courtyard-style planting verification area is a pre-selected courtyard area that is highly similar to the target courtyard in terms of spatial scale and environmental characteristics (shading, mixed greenery).

[0051] The courtyard remote sensing photography-specific abnormal parameter set is a collection of abnormal parameters extracted during the camera shooting process for courtyard remote sensing photography scenarios. This parameter set only includes abnormal parameters related to the courtyard shooting environment, such as exposure abnormalities caused by green plant occlusion, white balance abnormalities caused by wall reflections, and image blur abnormalities caused by small amplitude of drone vibrations. For example, the courtyard remote sensing photography-specific abnormal parameter set under a certain rotation azimuth angle is: exposure value offset +0.3, white balance color temperature offset 500K, and image blur 0.15.

[0052] The first anomaly parameter matrix is ​​a two-dimensional numerical matrix formed by arranging the specific anomaly parameter groups of the courtyard remote sensing images at each rotation azimuth angle in order of rotation azimuth angle. For example, the first anomaly parameter matrix arranged by rotation azimuth angles of 30°, 60°, and 90° is... The columns represent exposure value offset, white balance color temperature offset, and image blur, respectively, while the rows represent each rotation azimuth angle. The environmental disturbance conditions of the courtyard are the environmental factors that affect drone photography in the courtyard planting verification area. These include airflow disturbances, green plant obstruction, and wall reflections. For example, the environmental disturbance conditions of the courtyard at a certain rotation azimuth angle are: there is airflow disturbance in the center of the courtyard with a wind speed of 2 m / s, and there is strong light reflection on the west wall.

[0053] The second anomaly parameter matrix is ​​a two-dimensional numerical matrix formed by fine-tuning the anomaly parameters in the first anomaly parameter matrix. For example, the first anomaly parameter matrix... After fine-tuning, the second anomaly parameter matrix is ​​obtained. .

[0054] The first and second abnormal feature vectors are feature vectors obtained by performing eigenvalue decomposition on the first and second abnormal parameter matrices, respectively. These vectors reflect the core abnormal features of the matrix. For example, the first abnormal feature vector is [0.85, 0.32, 0.18] and the second abnormal feature vector is [0.42, 0.21, 0.10].

[0055] In this embodiment, the first / second abnormal parameter matrix is ​​decomposed into eigenvalues, and the contribution rate of each eigenvector is calculated as: contribution rate = (single eigenvalue / sum of all eigenvalues) × 100%. A contribution rate ≥ 20% is set as the judgment threshold. Abnormal parameters corresponding to eigenvectors with a contribution rate ≥ 20% are judged as non-persistent abnormal parameters (which can be corrected by adjusting camera parameters). Abnormal parameters corresponding to eigenvectors with a contribution rate < 20% are judged as persistent abnormal parameters. For example, non-persistent abnormal parameters include exposure shift, white balance shift, focal length shift, etc.

[0056] In this embodiment, the improvement coefficient = (arithmetic mean of the second column vector of the same non-stubborn anomaly parameter / arithmetic mean of the first column vector) × courtyard correction coefficient. The courtyard interference level corresponding to the courtyard correction coefficient (0.85~0.95) is determined by a comprehensive assessment of three quantitative indicators: green plant shading ratio, airflow speed, and wall reflectivity. The weights of each indicator are 0.5, 0.3, and 0.2, respectively. The specific judgment criteria are shown in Table 2. Table 2 Judgment Criteria Table

[0057] An improved sequence group is a sequence group formed by arranging the improvement coefficients of non-persistent anomaly parameters specific to remote sensing images of each courtyard according to parameter type. For example, an improved sequence group is [0.33, 0.6, 0.53], which corresponds to the improvement coefficients of exposure value shift, white balance color temperature shift, and image blur, respectively.

[0058] The difference sequence set is a set of quantitative values ​​of parameter differences obtained by comparing the improved sequence set with the non-persistent parameters in the standard parameters of a normal camera. This sequence set reflects the degree of difference between the calibrated camera parameters and the standard parameters.

[0059] In this embodiment, based on 100 sets of courtyard remote sensing shooting calibration experimental data, when the confidence level is ≥85%, the deviation of the image pixel resolution, focus clarity and other indicators of the drone camera is ≤15% compared with the standard camera, which can meet the feature extraction needs of small planting areas in the courtyard; if the confidence level is <85%, the image detail loss rate is >20%, and the poppy features cannot be accurately extracted, so the preset level is set to 85%.

[0060] In this embodiment, the attitude-condition-correction lookup table shows the correspondence between the attitude of the UAV's shooting amplitude, the interference conditions of the courtyard environment, and the correction conditions for abnormal parameters. The correction conditions target three types of non-persistent abnormal parameters in courtyard remote sensing photography: exposure shift, white balance shift, and image blur, as shown in Table 3. Table 3 Attitude-Condition-Correction Comparison Table

[0061] In this embodiment, the difference-camera comparison table is shown in Table 4: Table 4 Differences - Camera Comparison Table

[0062] The beneficial effects of the above technical solution are as follows: by selecting a courtyard-style planting verification area that is highly similar to the characteristics of the target courtyard, the camera calibration is adapted specifically to the courtyard scene. By constructing a unique abnormal parameter matrix for courtyard remote sensing photography and extracting feature vectors, non-stubborn abnormal parameters in the courtyard photography scene are accurately identified. By calculating the improvement coefficient and the difference sequence group, the accuracy of the camera shooting is quantitatively judged. The camera is replaced only when the confidence level of the difference sequence group is higher than 85%. This ensures that the shooting accuracy of the drone camera is adapted to the courtyard remote sensing photography environment and avoids unnecessary camera replacement, effectively improving the clarity and accuracy of subsequent courtyard remote sensing image shooting.

[0063] This invention provides a method for monitoring the illegal planting of poppies in a courtyard, comprising controlling a drone to capture several remote sensing images of the target courtyard, including: When the drone is shooting at the set shooting position, the drone's maximum amplitude and self-position are captured in real time, and the offset is compared with the current maximum amplitude and current position caused by the current shooting environment. If the contrast offset of the drone remains within the standard constraint range at all times during the shooting process, then the drone is controlled to shoot at the set shooting position according to the set number of shots. Otherwise, capture the instantaneous moments when the comparison offset is not within the standard constraint range, and determine the updated number of shots based on the time interval between the instantaneous moment and the start shooting moment, the extreme brewing variable based on the instantaneous moment and the current shooting accuracy of the drone. The extreme brewing variable is the sum of the drone amplitude change rate caused by the airflow disturbance of the courtyard building and the shooting field of view change rate caused by the occlusion of the courtyard green plants. The drone is controlled to continue taking pictures at the set shooting position according to the updated number of pictures.

[0064] In this embodiment, the comparison offset pair is obtained by comparing the maximum amplitude and self-position of the UAV with the current maximum amplitude and current position caused by the current shooting environment in the courtyard. The resulting two sets of offset data pairs include amplitude offset and positioning offset, such as amplitude offset pair: {5cm, 8cm}. The standard constraint range is the allowable range of the comparison offset pair set for courtyard remote sensing shooting, that is, the standard constraint range of amplitude offset is ≤0.2m, and the standard constraint range of positioning offset is ≤0.3m. At this time, the amplitude offset pair: {5cm, 8cm} satisfies the constraint. If the amplitude offset pair is {30cm, 40cm}, the occurrence time of the amplitude offset pair is regarded as the instantaneous time t1. If the shooting start time is t2, the time interval between the instantaneous time and the shooting start time is t1-t2. If only one of the amplitude offset or positioning offset is not within the corresponding standard constraint range, the change rate of the unsatisfied pair is calculated and the change rate of the satisfied pair is regarded as 0.

[0065] In this embodiment, the extreme incubation variable = ((actual amplitude of UAV - standard amplitude threshold) / standard amplitude) + ((100% - shooting field of view missing ratio) / 100%). This case applies when both amplitude offset and positioning offset are outside the corresponding standard constraint range, and the shooting field of view missing ratio = (theoretical shooting field of view area - actual shooting field of view area) / theoretical shooting field of view area.

[0066] When either the amplitude offset or the positioning offset is outside the corresponding standard constraint range, it is still necessary to obtain the corresponding instantaneous moment. In this case: If only the amplitude offset is outside the corresponding standard constraint range, the extreme incubation variable = (actual UAV amplitude - standard amplitude threshold) / standard amplitude; If only the positioning offset is outside the corresponding standard constraint range, the extreme incubation variable = (100% - the proportion of missing field of view) / 100%.

[0067] In this embodiment, N is the updated number of shots; N0 is the set number of shots; t0 is the maximum allowed shooting time per drone position; k is the courtyard correction coefficient; X is the extreme incubation variable; P is the shooting accuracy; ; The maximum shooting limit is set to 8; This is the upward value selection symbol.

[0068] Among them, based on the balance between efficiency and accuracy of courtyard remote sensing photography, when the number of shots is greater than 8, the image redundancy is greater than 70%, and the monitoring time increases by more than 50%. Therefore, the upper limit of shooting is set to 8 shots.

[0069] The beneficial effects of the above technical solution are as follows: by capturing the drone's comparative offset pairs in real time, dynamic monitoring of the drone's attitude and positioning during the courtyard shooting process is realized. By setting standard constraint ranges, it is possible to accurately determine whether the drone's shooting status meets the requirements. When the shooting status is abnormal, the number of shots is dynamically updated by combining the extreme brewing variables specific to the courtyard scene and the pixel coverage requirements of small planting areas, ensuring the image pixel coverage of small, occluded planting areas in the courtyard and avoiding the loss of image details caused by abnormal shooting attitude and positioning offset.

[0070] This invention provides a method for monitoring the illegal planting of poppies in courtyards, which constructs a combination sequence of corresponding one-dimensional vectors according to the spatial feature rules of courtyards, including: When the poppy matching element is an environment type, obtain the radiation space range of the matching element in the small planting areas such as flower pots, corners, and gaps between green plants in the target courtyard; When the matching element is an object type, the small-area object outline and object features of the matching element in the target courtyard are obtained, and the combination sequence is obtained by sequentially arranging the radiation space range and / or object outline and object features of the matching elements involved in the corresponding one-dimensional vector according to the courtyard space feature rules.

[0071] In this embodiment, the radiation space range refers to the spatial coordinate range of small planting areas such as flower pots, corners, and gaps between green plants within the courtyard. This range is presented in the form of latitude, longitude, and planar coordinates, reflecting the actual spatial location and area of ​​the planting area. For example, the radiation space range of the flower pot planting area in the center of the courtyard is: planar coordinates (X1:20, Y1:30) - (X2:25, Y2:35), and the area is... ; The object outlines include poppy seedlings and mature plants planted in a small area of ​​the courtyard, as well as exclusive object features adapted to the courtyard planting scene. The exclusive features of the seedlings include rosette-shaped plant shape and serrated leaves, while the object features include alternate lanceolate leaves and upright stems.

[0072] The beneficial effects of the above technical solution are: by accurately acquiring the radiation spatial range of the small planting area in the courtyard and the specific object characteristics of poppies, and constructing a combination sequence according to the spatial feature rules of the courtyard, the feature information of the combination sequence is highly matched with the actual spatial layout of the courtyard and the growth characteristics of poppies, avoiding the confusion of feature information of the combination sequence and improving the accuracy of the subsequent integration of the combination sequence into the second image.

[0073] This invention provides a method for monitoring the illegal planting of poppies in courtyards, comprising determining a set of courtyard feature comparison coefficients between each single-dimensional vector in the same remote sensing image and the stitched vector, including: The single-dimensional vector and the spliced ​​vector are normalized and aligned to reflect the dimensions of poppy monitoring-related features in the courtyard. The missing poppy monitoring-related feature dimensions in the single-dimensional vector are padded with zero values ​​to obtain the aligned single-dimensional vector and the aligned spliced ​​vector. The feature dimensions corresponding to the standard environment and standard growth of poppies in different growth cycles under the courtyard scene in the spliced ​​vector are extracted as core dimensions. The core dimensions are assigned a first weight, and the other non-core dimensions are assigned a second weight, with the first weight being greater than the second weight. Calculate the cosine similarity coefficient and Euclidean distance deviation coefficient of the aligned single-dimensional vector and the aligned concatenated vector in each dimension respectively. Combine the weights of the corresponding dimensions to perform a weighted fusion operation on the two coefficients to obtain the fusion coefficients of each dimension. The fusion coefficients of all dimensions are combined according to the order of the dimension types of the yard poppy monitoring to obtain the control coefficient group corresponding to each single-dimensional vector and the spliced ​​vector. Each fusion coefficient in the control coefficient group is associated with the dimension type and weight assignment information of the corresponding yard poppy monitoring.

[0074] In this embodiment, if the dimensions of the splicing vector are: [leaf outline, plant shape, color, planting area], and the dimensions of a certain single-dimensional vector are: [leaf outline, color], after padding the zero-value vector, it becomes: [leaf outline, 0, color, 0], which is aligned with the dimensions of the splicing vector.

[0075] In this embodiment, the core dimensions are the feature dimensions that are strongly correlated with poppy identification in the monitoring of poppies in the courtyard. For example, the core dimensions of poppy seedlings are leaf outline, plant height, and flower pot planting space features. The core dimensions of mature poppy plants are leaf outline, plant height, petals, and other non-core dimensions such as the tile material of the courtyard ground and the white color of the fence.

[0076] Fusion coefficient = cosine similarity coefficient × corresponding dimension weight + Euclidean distance deviation coefficient × (1 - corresponding dimension weight).

[0077] The beneficial effects of the above technical solution are as follows: By aligning the single-dimensional vector and the spliced ​​vector through dimension normalization, the accuracy of vector calculation is ensured; by dividing the feature dimension into core and non-core dimensions and assigning differentiated weights, the feature information strongly related to poppy identification is highlighted, improving the accuracy of feature matching; by combining the cosine similarity coefficient and the Euclidean distance deviation coefficient to calculate the fusion coefficient, the feature matching degree and deviation degree of the vector are comprehensively reflected, so that the calculated courtyard feature comparison coefficient group can accurately reflect the actual matching situation of the single-dimensional vector and the spliced ​​vector, effectively reducing the misjudgment rate of poppy matching elements and improving the accuracy of courtyard poppy feature matching.

[0078] This invention provides a method for monitoring the illegal planting of poppies in courtyards. It involves integrating all combined sequences assigned comprehensive analysis labels according to the mapping relationship of dimensional types to obtain a second image, including: By classifying the combined sequences with comprehensive analysis labels under the same remote sensing image, we can obtain several categories of combined elements; The dimension type with the highest mapping confidence is extracted from the mapping relationship of the corresponding dimension type of the comprehensive analysis label and regarded as the first dimension. The element features based on the first dimension are then used to construct the first contour baseline. Meanwhile, the dimension type that is closest to the feature contour is extracted from the dimension type assigned to the comprehensive analysis label and regarded as the second dimension. At this time, the second contour baseline is obtained based on the element features of the second dimension. The first contour baseline and the second contour baseline are aligned. When the first contour baseline and the second contour baseline are completely consistent, the second contour baseline is used as the initial contour baseline. Otherwise, lock the inconsistency point of the first contour baseline. When the inconsistency point is located inside the second contour baseline, determine the first nearest perpendicular point between the inconsistency point and the second contour baseline, and lock the center point of the straight line connecting the first nearest perpendicular point and the inconsistency point. At the same time, obtain the two second adjacent points through which the tangent of the inconsistency point passes through the second contour baseline, and the curvature direction of the inconsistency point in the second contour baseline. Based on the ratio of the first length between the two adjacent points to the baseline length of the first contour baseline located inside the second contour baseline, and in conjunction with the curvature direction and the angle between the curvature direction and the connecting line, the center point is adjusted along the connecting line. When the inconsistency point is located outside the second contour baseline, the average of the nearest distance difference between all inconsistency points located outside the second contour baseline and the second contour baseline is obtained. At the same time, the baseline length between the corresponding inconsistency point and the second contour baseline and the nearest coincident point on the second contour baseline is obtained and regarded as the first length. The inconsistency points are adjusted a second time along a line perpendicular to the second profile baseline based on the average value, the shortest distance length, and the first length. A preliminary contour baseline is obtained by integrating the adjusted points and the unadjusted points; For each type of remaining combination element, supplementary features for monitoring poppies in the courtyard are extracted according to the corresponding dimension type. In addition, the corresponding supplementary features are preferentially superimposed with the poppy feature parameters of the courtyard shading area for radiation processing, which is applied within the framework of the preliminary contour baseline. When all remaining combination elements have been radiated, the second image is obtained.

[0079] In this embodiment, the maximum mapping confidence is the maximum value of the correlation between the dimension type and the poppy monitoring feature, which is extracted from the mapping relationship of the dimension type. That is, it can be directly obtained from the table that stores the mapping confidence values ​​of each dimension type and poppy monitoring in advance.

[0080] The first contour baseline is a preliminary contour baseline constructed based on the elemental features of the first dimension and the standard contour features of the poppy. For example, the closed circular contour baseline constructed based on the elemental features of the first dimension contour and combined with the standard rosette-shaped contour of the poppy seedling is the first contour baseline.

[0081] In this embodiment, the similarity between the features of each dimension type and the standard outline features of poppies is calculated, and the dimension type with the highest similarity is marked as the second dimension. At the same time, the element features under this dimension are retrieved.

[0082] The second contour baseline is a contour baseline constructed based on the element features of the second dimension and combined with the standard contour features of poppies. This baseline is used to compare with the first contour baseline and supplement and improve the detailed information of the poppy contour. For example, based on the element features of the second dimension color dimension and combined with the color distribution features of poppy seedling leaves, the contour baseline that matches the shape of the first contour baseline is the second contour baseline.

[0083] Position alignment processing is a method of matching the coordinate origin and feature key points of the first contour baseline and the second contour baseline. The purpose is to make the two contour baselines compare in the same coordinate system and ensure the accurate determination of inconsistencies in the contours.

[0084] Inconsistency points are feature points where the coordinates of the first and second contour baselines do not coincide after the alignment process. These points are the core objects of contour baseline adjustment and are often caused by the obstruction of garden greenery or the deviation of the shooting angle. For example, in the two contour baselines after alignment, a certain coordinate point of the first contour baseline is (10,15), and the corresponding coordinate point of the second contour baseline is (12,16). This point is the inconsistency point.

[0085] For example, the midpoint of the line connecting the inconsistent point (8,9) and the first nearest perpendicular point (8,10) is (8,9.5).

[0086] The tangent is a straight line drawn through the inconsistency point and in the same direction as the curvature of the second contour baseline. For example, if an arc-shaped tangent is drawn through the inconsistency point (8,9) inside the second contour baseline, and the curvature of the rosette-shaped contour at that position is consistent, the straight line is the required tangent.

[0087] The second nearest point is the two intersections formed by the tangent of the inconsistency point and the second contour baseline. For example, the intersections formed by the tangent of the inconsistency point and the second contour baseline are (5,12) and (11,12). These two points are the second nearest points.

[0088] The curvature direction is the direction in which the curve of the second profile baseline bends at the inconsistency point. It includes two types: clockwise curvature and counterclockwise curvature. This direction is the basis for the first adjustment of the inconsistency point. For example, if the second profile baseline bends clockwise at the inconsistency point (8,9), then the curvature direction of that point is clockwise.

[0089] The average of the closest distance differences is calculated by taking the nearest distance from each inconsistency point to the second profile baseline for all inconsistencies located outside the second profile baseline, and then taking the arithmetic mean of all distance differences.

[0090] The shortest distance length is the vertical distance from a single inconsistency point located outside the second contour baseline to the second contour baseline. For example, the vertical distance from the external inconsistency point (15,20) to the second contour baseline is 1.2 pixels, and this value is the shortest distance length.

[0091] The remaining combination elements are the combination elements other than those used to construct the first and second contour baselines after the same combination elements have been classified.

[0092] Supplementary features are detailed features related to poppy monitoring extracted from the remaining combined elements according to the corresponding dimension type. These features are used to enrich the internal details of the preliminary outline baseline and improve the integrity of the outline. For example, the serrated texture feature of poppy leaves extracted from the remaining combined elements of the texture dimension and the rosette-shaped layered feature of seedlings extracted from the plant shape dimension are both supplementary features.

[0093] In this embodiment, the first adjustment is: Where L1 is the first length; L2 is the inner baseline length; and θ is the included angle. This is the curvature direction correction coefficient, with a value of 1 for clockwise bending and a value of -1 for counterclockwise bending; The basic adjustment coefficient is 0.05~0.2 (0.05~0.1 for seedlings and 0.1~0.2 for mature plants). Adjust the amount for the center point; Where (x01, y01) are the original coordinates of the center point; (x1, y1) are the adjusted coordinates of the center point; The direction vector of the connecting line is the unit direction vector of the line connecting the center point and the first nearest perpendicular point; In this embodiment, the second adjustment is: , The baseline length of the coincident point; This is the total length of the second profile baseline; This is the baseline length correction factor; , Adjustment amount for inconsistencies; This is the distance correction factor; The shortest distance; This is the average value; (x02, y02) are the original coordinates of the inconsistency point; (px1, py1) are the coordinates of the inconsistency point after adjustment. The normal vector of the vertical line is the unit normal vector of the second profile baseline at the nearest position to the point of inconsistency.

[0094] The beneficial effects of the above technical solution are as follows: By classifying and filtering the core dimension types according to the combination elements to construct a dual contour baseline, the dual-layer positioning of poppy contour features is achieved. In combination with the problem of contour incompleteness caused by courtyard greenery, a special inconsistency adjustment rule is designed. Through the first and second adjustments, the contour baseline is made to accurately fit the actual shape of poppies in the courtyard scene. Then, the contour details are enriched by the supplementary feature radiation processing of the remaining combination elements, so that the integrated second image can completely restore the core features and details of poppies in the courtyard occlusion environment. This solves the problem of poppy contour incompleteness and feature loss in the existing technology, improves the completeness of poppy features in the second image, and further effectively improves the accuracy of poppy recognition.

[0095] This invention provides a method for monitoring the illegal cultivation of poppies in courtyards, which involves applying radiation treatment to corresponding supplementary features within the framework of the preliminary contour baseline, including: The mapping similarity between the remaining combined elements of the courtyard poppy monitoring dimension type and the first and second dimensions is calculated based on the cosine similarity algorithm, and then matched with the level comparison table to obtain the mapping level. When the mapping level is high priority radiation, the leaf contour and plant height parameters that are strongly correlated with the monitoring of poppies in the courtyard are weighted and superimposed onto the core area of ​​the preliminary contour baseline. When the mapping level is medium priority radiation, the corresponding supplementary features are scaled and normalized, and pixel-level weighted fusion is performed to inject the secondary core region of the preliminary contour baseline. When the mapping level is low-priority radiation, the corresponding supplementary features are used as background enhancement information and low-intensity superimposed on the edge transition region of the initial contour baseline.

[0096] In this embodiment, the mapping similarity is the feature association quantification value between the dimension type of the remaining combined elements and the first and second dimensions, calculated based on the cosine similarity algorithm.

[0097] In this embodiment, the grade comparison table is a pre-constructed comparison table that corresponds one-to-one with the mapping similarity value and the radiation treatment level. This table is specifically for monitoring poppies in the yard and divides the mapping similarity into different intervals, corresponding to different radiation priorities. For example, the grade comparison table stipulates that mapping similarity ≥ 0.9 is high priority radiation, 0.7 ≤ mapping similarity < 0.9 is medium priority radiation, and mapping similarity < 0.7 is low priority radiation.

[0098] In this embodiment, the core quantitative parameters that directly affect poppy identification are extracted from the supplementary features. These parameters are key features that distinguish poppies from other green plants, such as the number of serrations on poppy leaves, the spacing between serrations, the aspect ratio of leaf length to width, and the ratio of plant height to crown width. These are all parameters that are strongly correlated with poppy monitoring.

[0099] The core region is the key feature area of ​​the poppy within the preliminary outline baseline. This region is the core part for poppy identification, such as the leaf area, flower stalk area, and the central area of ​​the rosette-shaped plant.

[0100] Medium-priority radiation is a supplementary feature that is moderately relevant to poppy monitoring. After scale normalization, it is incorporated into the secondary core region in a pixel-level weighted fusion manner. For example, the color gradient feature of poppy leaves is a moderately relevant feature. Scale normalization is a process that adjusts the pixel size and numerical range of supplementary features to a standard range that matches the initial contour baseline. The purpose is to make the scale of the supplementary features consistent with the contour baseline and avoid feature distortion during fusion. For example, adjusting the pixel size of the supplementary features from 20×20 to 50×50, which matches the core area of ​​the initial contour baseline, completes the scale normalization.

[0101] Pixel-level weighted fusion is a process that weights the scale-normalized supplementary features and the secondary core region of the initial contour baseline by pixels. For example, the pixel values ​​of the color gradient supplementary features and the pixel values ​​of the secondary core region are weighted by 0.7 and 0.3 respectively to complete the pixel-level weighted fusion.

[0102] The secondary core region is the peripheral feature area surrounding the core region within the initial outline baseline. This region supplements the core region, such as the leaf edge area of ​​a poppy seedling or the outer area of ​​the plant shape. For example, the rosette-shaped outer area outside the core leaf area of ​​a poppy seedling is the secondary core region.

[0103] Background enhancement information is weakly correlated with poppy monitoring and is supplementary information used to enrich the background features of images. This type of information does not affect the identification of core poppy features and is only used to improve the scene integrity of the image. For example, the soil texture in the flower pot planting area of ​​the courtyard and the texture of the flower pot edge are all background enhancement information.

[0104] The edge transition region is the outer edge region of the initial outline baseline. This region is the transition area between the poppy outline and the courtyard background. For example, the ring-shaped area 5 pixels outside the initial outline baseline of the poppy seedling is the edge transition region.

[0105] Low-intensity overlay is a processing method that overlays background enhancement information onto the edge transition area with low weight and low transparency. For example, the background enhancement information of soil texture is overlaid onto the edge transition area with a weight of 0.2 to complete the low-intensity overlay.

[0106] The beneficial effects of the above technical solution are as follows: Based on the cosine similarity algorithm, the mapping similarity is calculated and the radiation level is divided. Different radiation processing methods and superposition areas are designed for the supplementary features of different levels. This achieves the layered and accurate fusion of supplementary features and the initial contour baseline, making strong correlation features stand out in the core area, medium correlation features supplement the secondary core area, and weak correlation features enrich the edge area as background. This not only ensures the prominence of the poppy core identification features, but also improves the scene integrity of the image, and solves the problems of chaotic feature fusion and the obscuring of core features in the existing technology.

[0107] This invention provides a monitoring system for illegal poppy cultivation in courtyards, such as... Figure 2 As shown, it includes: The reconstruction and stitching module is used to control the drone to capture several remote sensing images of the target courtyard and perform several single-dimensional reconstruction processes on each remote sensing image to obtain a one-to-one corresponding single-dimensional image. At the same time, environmental features and existing object features are extracted from each remote sensing image, and compared with the standard environment for planting poppies in the courtyard scene and the standard growth of different growth cycles to obtain the stitching vector of the corresponding remote sensing image with courtyard scene feature weights. Among them, there is at least one remote sensing image under different set shooting positions. The single-dimensional decomposition module is used to match the image decomposition method from the type-image analysis lookup table according to the dimension type of the single dimension, and to decompose the corresponding single-dimensional image according to the image decomposition method to obtain the single-dimensional vector for monitoring the poppy in the courtyard. The element judgment module is used to determine the courtyard feature comparison coefficient group of each single-dimensional vector under the same remote sensing image and the stitched vector, and to determine whether there is a poppy matching element in the corresponding single-dimensional vector. If there is, the combination sequence of the corresponding single-dimensional vector is constructed according to the courtyard spatial feature rules and a comprehensive analysis label is assigned. If there is no poppy matching element, an independent analysis label is assigned to the corresponding single-dimensional vector. The image annotation module is used to input a single-dimensional image of a single-dimensional vector with an independent analysis label into a general poppy image recognition model and output a first image with annotations. At the same time, all combination sequences with comprehensive analysis labels are integrated in detail according to the mapping relationship of dimension type to obtain a second image. The second image is then input into a poppy recognition model for different growth stages of poppies in a courtyard scene and outputs a second image with annotations in sequence. The monitoring integration module is used to perform redundant processing on all labeled images from the same shooting location, and to perform feature weighted fusion processing on the redundant images of the courtyard scene to obtain monitoring results of illegal poppy planting in the target courtyard, including planting location, area, and growth cycle.

[0108] The beneficial effects of the above technical solution are as follows: By setting drone shooting parameters specific to courtyards, clear image acquisition of small-area, obscured planting areas in courtyards is ensured; by constructing feature vectors based on courtyard scene feature weights and designing a specific set of control coefficients, the accuracy of poppy feature matching is improved; through a dual-model recognition system of initial screening using a general model and fine screening using a courtyard-specific growth cycle model, the false negative and false positive rates of courtyard poppy monitoring are reduced; and finally, through image redundancy processing and feature weighted fusion specific to courtyard scenes, the comprehensiveness and accuracy of monitoring results are achieved, significantly improving the accuracy and efficiency of monitoring small-area, obscured illegal poppy cultivation in courtyards.

[0109] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for monitoring the illegal cultivation of poppies in a courtyard, characterized in that, include: Step 1: Control the drone to capture several remote sensing images of the target courtyard and perform several single-dimensional reconstruction processes on each remote sensing image to obtain a one-to-one corresponding single-dimensional image. At the same time, extract environmental features and existing object features from each remote sensing image and compare it with the standard environment for planting poppies in the courtyard scene and the standard growth of different growth cycles to obtain the stitched vector of the corresponding remote sensing image with courtyard scene feature weights. Among them, there is at least one remote sensing image under different set shooting positions. Step 2: Match the image decomposition method from the type-image analysis lookup table according to the dimension type of the single dimension, and decompose the corresponding single-dimensional image according to the image decomposition method to obtain the single-dimensional vector for monitoring poppies in the courtyard; Step 3: Determine the courtyard feature comparison coefficient group of each single-dimensional vector under the same remote sensing image and the stitched vector respectively, and determine whether there is a poppy matching element in the corresponding single-dimensional vector. If there is, construct the combination sequence of the corresponding single-dimensional vector according to the courtyard spatial feature rules and assign a comprehensive analysis label. If there is no poppy matching element, assign an independent analysis label to the corresponding single-dimensional vector. Step 4: Input the single-dimensional image of the single-dimensional vector with independent analysis labels into the general poppy image recognition model, and output the first image with labels. At the same time, integrate all the combination sequences with comprehensive analysis labels according to the mapping relationship of dimension type to obtain the second image. Then, input the second image into the poppy recognition model with different growth stages of poppies in the courtyard scene in turn, and output the second image with labels in turn. Step 5: Perform redundancy processing on all labeled images from the same shooting location, and then perform feature weighted fusion processing on the redundant images to obtain the monitoring results of illegal poppy planting in the target courtyard, including planting location, area, and growth cycle.

2. The method for monitoring illegal poppy cultivation in courtyards according to claim 1, characterized in that, Before using remote sensing images of the target courtyard taken by drone, the following steps are included: The drones are controlled to take pictures of the courtyard-style planting verification area, which is covered by walls and green plants, in turn at a rotation azimuth angle of 30°. The images to be analyzed are obtained at each rotation azimuth angle. At the same time, the standard images of the courtyard-style planting verification area taken by the normal camera at each rotation azimuth angle are obtained. The image to be analyzed at the same rotation azimuth angle is compared with the standard image at the same location point to obtain the first restored image by proportionally restoring the corresponding image to be analyzed. The courtyard remote sensing-specific abnormal parameter group of each first restored image is obtained and the first abnormal parameter matrix is ​​constructed. The shooting amplitude attitude and courtyard environment interference conditions of the UAV at each rotation azimuth angle are retrieved and matched with the attitude-condition-correction lookup table to obtain correction conditions. Each abnormal parameter in the courtyard remote sensing shooting exclusive abnormal parameter group of the first restored image at the corresponding rotation azimuth angle is fine-tuned. Then, a second abnormal parameter matrix is ​​constructed based on the combination of all fine-tuned parameters. Obtain the first and second anomalous feature vectors of the first and second anomalous parameter matrices, and determine the non-persistent anomalous parameters of the courtyard remote sensing images. Extract the first column vector and the second column vector of each non-stubborn anomaly parameter from the first anomaly parameter matrix and the second anomaly parameter matrix respectively, and calculate the improvement coefficient of the corresponding non-stubborn anomaly parameter based on the first column vector and the second column vector of the same non-stubborn anomaly parameter to obtain an improved sequence group including the non-stubborn anomaly parameter. The improved sequence set is compared with the non-persistent parameters in the standard parameters of the normal camera to obtain the difference sequence set; When the confidence level of the difference sequence group is greater than a preset level, the original camera installed on the drone is replaced by matching the camera to be replaced from the difference-camera lookup table that matches the difference sequence group. Otherwise, it is determined that the original camera installed on the drone does not need to be replaced.

3. The method for monitoring illegal poppy cultivation in courtyards according to claim 1, characterized in that, Control the drone to capture several remote sensing images of the target courtyard, including: When the drone is shooting at the set shooting position, the drone's maximum amplitude and self-position are captured in real time, and the offset is compared with the current maximum amplitude and current position caused by the current shooting environment. If the contrast offset of the drone remains within the standard constraint range at all times during the shooting process, then the drone is controlled to shoot at the set shooting position according to the set number of shots. Otherwise, capture the instantaneous moments when the comparison offset is not within the standard constraint range, and determine the updated number of shots based on the time interval between the instantaneous moment and the start shooting moment, the extreme brewing variable based on the instantaneous moment and the current shooting accuracy of the drone. The extreme brewing variable is the sum of the drone amplitude change rate caused by the airflow disturbance of the courtyard building and the shooting field of view change rate caused by the occlusion of the courtyard green plants. The drone is controlled to continue taking pictures at the set shooting position according to the updated number of pictures.

4. The method for monitoring illegal poppy cultivation in courtyards according to claim 1, characterized in that, Construct a combination sequence of corresponding one-dimensional vectors according to the spatial feature rules of the courtyard, including: When the poppy matching element is an environment type, obtain the radiation space range of the matching element in the small planting areas such as flower pots, corners, and gaps between green plants in the target courtyard; When the matching element is an object type, the small-area object outline and object features of the matching element in the target courtyard are obtained, and the combination sequence is obtained by sequentially arranging the radiation space range and / or object outline and object features of the matching elements involved in the corresponding one-dimensional vector according to the courtyard space feature rules.

5. The method for monitoring illegal poppy cultivation in courtyards according to claim 1, characterized in that, Determine the courtyard feature comparison coefficient set for each single-dimensional vector in the same remote sensing image and the stitched vector, including: The single-dimensional vector and the spliced ​​vector are normalized and aligned to reflect the dimensions of poppy monitoring-related features in the courtyard. The missing poppy monitoring-related feature dimensions in the single-dimensional vector are padded with zero values ​​to obtain the aligned single-dimensional vector and the aligned spliced ​​vector. The feature dimensions corresponding to the standard environment and standard growth of poppies in different growth cycles under the courtyard scene in the spliced ​​vector are extracted as core dimensions. The core dimensions are assigned a first weight, and the other non-core dimensions are assigned a second weight, with the first weight being greater than the second weight. Calculate the cosine similarity coefficient and Euclidean distance deviation coefficient of the aligned single-dimensional vector and the aligned concatenated vector in each dimension respectively. Combine the weights of the corresponding dimensions to perform a weighted fusion operation on the two coefficients to obtain the fusion coefficients of each dimension. The fusion coefficients of all dimensions are combined according to the order of the dimension types of the yard poppy monitoring to obtain the control coefficient group corresponding to each single-dimensional vector and the spliced ​​vector. Each fusion coefficient in the control coefficient group is associated with the dimension type and weight assignment information of the corresponding yard poppy monitoring.

6. The method for monitoring illegal poppy cultivation in courtyards according to claim 1, characterized in that, The second image is obtained by integrating all combined sequences labeled with comprehensive analysis according to the mapping relationship of dimension type, including: By classifying the combined sequences with comprehensive analysis labels under the same remote sensing image, we can obtain several categories of combined elements; The dimension type with the highest mapping confidence is extracted from the mapping relationship of the corresponding dimension type of the comprehensive analysis label and regarded as the first dimension. The element features based on the first dimension are then used to construct the first contour baseline. Meanwhile, the dimension type that is closest to the feature contour is extracted from the dimension type assigned to the comprehensive analysis label and regarded as the second dimension. At this time, the second contour baseline is obtained based on the element features of the second dimension. The first contour baseline and the second contour baseline are aligned. When the first contour baseline and the second contour baseline are completely consistent, the second contour baseline is used as the initial contour baseline. Otherwise, lock the inconsistency point of the first contour baseline. When the inconsistency point is located inside the second contour baseline, determine the first nearest perpendicular point between the inconsistency point and the second contour baseline, and lock the center point of the straight line connecting the first nearest perpendicular point and the inconsistency point. At the same time, obtain the two second adjacent points through which the tangent of the inconsistency point passes through the second contour baseline, and the curvature direction of the inconsistency point in the second contour baseline. Based on the ratio of the first length between the two adjacent points to the baseline length of the first contour baseline located inside the second contour baseline, and in conjunction with the curvature direction and the angle between the curvature direction and the connecting line, the center point is adjusted along the connecting line. When the inconsistency point is located outside the second contour baseline, the average of the nearest distance difference between all inconsistency points located outside the second contour baseline and the second contour baseline is obtained. At the same time, the baseline length between the corresponding inconsistency point and the second contour baseline and the nearest coincident point on the second contour baseline is obtained and regarded as the first length. The inconsistency points are adjusted a second time along a line perpendicular to the second profile baseline based on the average value, the shortest distance length, and the first length. A preliminary contour baseline is obtained by integrating the adjusted points and the unadjusted points; For each type of remaining combination element, supplementary features for monitoring poppies in the courtyard are extracted according to the corresponding dimension type. In addition, the corresponding supplementary features are preferentially superimposed with the poppy feature parameters of the courtyard shading area for radiation processing, which is applied within the framework of the preliminary contour baseline. When all remaining combination elements have been radiated, the second image is obtained.

7. The method for monitoring illegal poppy cultivation in courtyards according to claim 5, characterized in that, Radiation processing of the corresponding supplementary features is applied within the framework of the initial contour baseline, including: The mapping similarity between the remaining combined elements of the courtyard poppy monitoring dimension type and the first and second dimensions is calculated based on the cosine similarity algorithm, and then matched with the level comparison table to obtain the mapping level. When the mapping level is high priority radiation, the leaf contour and plant height parameters that are strongly correlated with the monitoring of poppies in the courtyard are weighted and superimposed onto the core area of ​​the preliminary contour baseline. When the mapping level is medium priority radiation, the corresponding supplementary features are scaled and normalized, and pixel-level weighted fusion is performed to inject the secondary core region of the preliminary contour baseline. When the mapping level is low-priority radiation, the corresponding supplementary features are used as background enhancement information and low-intensity superimposed on the edge transition region of the initial contour baseline.

8. A monitoring system for illegal poppy cultivation in courtyards, characterized in that, include: The reconstruction and stitching module is used to control the drone to capture several remote sensing images of the target courtyard and perform several single-dimensional reconstruction processes on each remote sensing image to obtain a one-to-one corresponding single-dimensional image. At the same time, environmental features and existing object features are extracted from each remote sensing image, and compared with the standard environment for planting poppies in the courtyard scene and the standard growth of different growth cycles to obtain the stitching vector of the corresponding remote sensing image with courtyard scene feature weights. Among them, there is at least one remote sensing image under different set shooting positions. The single-dimensional decomposition module is used to match the image decomposition method from the type-image analysis lookup table according to the dimension type of the single dimension, and to decompose the corresponding single-dimensional image according to the image decomposition method to obtain the single-dimensional vector for monitoring the poppy in the courtyard. The element judgment module is used to determine the courtyard feature comparison coefficient group of each single-dimensional vector under the same remote sensing image and the stitched vector, and to determine whether there is a poppy matching element in the corresponding single-dimensional vector. If there is, the combination sequence of the corresponding single-dimensional vector is constructed according to the courtyard spatial feature rules and a comprehensive analysis label is assigned. If there is no poppy matching element, an independent analysis label is assigned to the corresponding single-dimensional vector. The image annotation module is used to input a single-dimensional image of a single-dimensional vector with an independent analysis label into a general poppy image recognition model and output a first image with annotations. At the same time, all combination sequences with comprehensive analysis labels are integrated in detail according to the mapping relationship of dimension type to obtain a second image. The second image is then input into a poppy recognition model for different growth stages of poppies in a courtyard scene and outputs a second image with annotations in sequence. The monitoring integration module is used to perform redundant processing on all labeled images from the same shooting location, and to perform feature weighted fusion processing on the redundant images of the courtyard scene to obtain monitoring results of illegal poppy planting in the target courtyard, including planting location, area, and growth cycle.

Images