Unmanned aerial vehicle-based termite exposed feature intelligent detection method and termite nest positioning method

Abstract

The application relates to a kind of termite exposed feature intelligent detection method and nest positioning method based on unmanned aerial vehicle, the termite exposed feature intelligent detection method is by pre-constructed termite exposed feature image feature parameter set representing termite exposed feature, and based on termite exposed feature image feature parameter set, termite exposed feature detection model is constructed, and then the image to be detected obtained by unmanned aerial vehicle in the region to be detected and after processing is input into termite exposed feature detection model, to judge whether the image to be detected contains termite exposed feature based on the detection result of the termite exposed feature detection model, and further based on the detection result of termite exposed feature, it is determined whether there is termite activity in the region to be detected, the intelligent detection of termite exposed feature for the region to be detected is realized, which is beneficial to improve the calibration efficiency and accuracy of the position of termite exposed feature in the region to be detected.

Description

Technical Field

[0001] This invention relates to the field of termite detection, and in particular to an intelligent detection method for exposed termite features and a method for termite nest location based on unmanned aerial vehicles (UAVs). Background Technology

[0002] Termites typically live in environments where there is no light or wind, making it difficult for the human eye to directly observe their activities, as well as the size and location of their nests.

[0003] To determine the activity of termites in dikes or forest areas, especially their location, current termite detection methods typically employ manual surveying and marking. That is, personnel go to the dikes or forest areas to inspect the damage and activity of soil-dwelling termites within the dikes or forest areas. Based on the exposed characteristics of termite activity within the dikes or forest areas (exposed termite characteristics include surface symbols of termite activity and termite nest indicators; surface symbols include termite trails and mud cover, and termite nest indicators include umbrella-shaped burrows, anthracnose fungi, swarming holes, ventilation holes, excrement, and termite mounds), the locations of the exposed termite characteristics are marked. Then, based on the determined locations of the exposed termite characteristics, the location of the termite nest is manually determined. Finally, the locations of the exposed termite characteristics and the nest are recorded in the termite marking record material. When a re-inspection is needed, personnel go to the forest area and arrive at the locations of the exposed termite characteristics and the nest recorded in the previous termite marking record material to understand the termite activity during this period, such as whether the termite damage has been controlled.

[0004] However, existing methods for detecting exposed termite features have shortcomings: because these exposed termite features are attached to the ground surface, they are easily destroyed by external factors such as animal activity, human activity, or natural wind. For example, termite trails, mud covers, swarming holes, ventilation holes, water lines, and fungal indicators are easily destroyed by biological activity or wind and rain. This causes some exposed termite features to be missed during the inspection process. Even physical markings of exposed termite features discovered during the inspection process can disappear due to external factors such as animal activity, human activity, or natural wind. This makes it impossible to determine the location of previously marked exposed termite features during re-inspection or verification, which adversely affects the location of termite indicators, the location of termite nests, and the understanding of termite activity. Summary of the Invention

[0005] The first technical problem to be solved by the present invention is to provide an intelligent detection method for exposed termite features based on drones, which is in contrast to the above-mentioned prior art.

[0006] The second technical problem to be solved by this invention is to provide a method for locating termite nests, which addresses the aforementioned prior art. This nest location method utilizes the aforementioned intelligent detection method for exposed termite features.

[0007] The technical solution adopted by this invention to solve the first technical problem is: an intelligent detection method for termite exposure features based on unmanned aerial vehicles (UAVs), characterized by comprising the following steps:

[0008] Step S1: Pre-construct a set of image feature parameters representing the exposed features of termites; wherein the set of image feature parameters representing the exposed features of termites includes at least one image feature parameter representing the exposed features of termites.

[0009] Step S2: Construct a termite exposure feature detection model based on the feature parameter set of termite exposure feature images;

[0010] Step S3: Use a drone to acquire sample images of the area to be detected;

[0011] Step S4: Process the sample image to obtain the image to be detected;

[0012] Step S5: Input the image to be detected into the termite exposure feature detection model to obtain the termite exposure feature detection result for the image to be detected.

[0013] Improved, in the aforementioned UAV-based intelligent detection method for termite exposure features, the pre-construction process of the termite exposure feature image feature parameter set is as follows:

[0014] Step a1: Predetermine the set of target image feature parameters to be detected when performing termite exposure feature detection on the sample images acquired by the UAV;

[0015] Step a2: Based on the determined target image feature parameter set, construct a termite exposure feature image feature parameter set for identifying the termite exposure features.

[0016] Furthermore, in the aforementioned UAV-based intelligent detection method for termite exposure features, the construction process of the termite exposure feature detection model is as follows:

[0017] Step b1: Use a convolutional neural network to obtain a feature map of termite exposure features in the image to be detected; wherein, the feature map of termite exposure features is labeled as f. map_h ,:

[0018]

[0019] Where CNN(·) is the forward propagation function of the selected convolutional neural network CNN, x img Let H be the image to be detected, and let H be the image to be detected x.img The length of the image x to be detected is W, where W is the length of the image x to be detected. img The width of C1 represents the number of channels in the feature map output by the forward propagation function CNN(·).

[0020] Step b2: Perform dimensionality reduction on the termite exposure feature image to obtain a dimensionality-reduced feature map; wherein, the dimensionality-reduced feature map after the dimensionality reduction operation is labeled as f. map :

[0021] f map =conv 1×1 (f map_h ), f map ∈R C×H×W ;

[0022] Among them, f map Image feature map f representing exposed termite characteristics map_h The dimensionality-reduced feature map after dimensionality reduction operation, conv 1×1 (·) represents a convolution operation with a kernel size of 1 and a number of channels of C, where C represents the number of channels in the reduced feature map;

[0023] Step b3: Perform feature dimensionality reduction on the obtained dimensionality-reduced feature map to obtain the dimensionality-reduced feature vector, and construct the image feature extraction backbone model using this feature vector; wherein, the dimensionality-reduced feature vector is labeled f, and the image feature extraction backbone model is labeled IMGEXT(·):

[0024] f = σ(W1, (pad(f)) map ))), f∈R C×D ;

[0025] Where σ(·) represents the activation function of the feedforward neural network, W1 represents the learnable parameter matrix of the feedforward neural network, and pad(f map ) represents flattening a feature map of dimension H×W into a one-dimensional vector, where D is the dimension of the feature vector f;

[0026] Step b4: Based on the dimensionality-reduced feature vectors obtained from the features, construct a target detection feature extraction model; wherein, the target detection feature extraction model is denoted as LOCEXT(·):

[0027] f i+1 =Transformer i (f i ),f i ∈R C×D f1 = f, f i+1 ∈R C×D , i∈N+Λi≤L;

[0028] Among them, f iLet i represent the detection feature of the i-th object, and let Transformer be the transform. i (·) represents the forward propagation process of the i-th layer self-attention network Transformer, f L This represents the final target detection features;

[0029] Step b5: Based on the obtained final target detection features, construct a regression prediction network to predict the specific location of the termite exposure features within the image to be detected;

[0030] f b o x =max(0,f L W2+b2)W3+b3,f box ∈R C×5 ;

[0031] Where max(0,f) L W2+b2) represents the relationship between the numerical values ​​0 and f. L The maximum value is taken from W2+b2, where W2, W3, b2, and b3 are learnable parameters in the feedforward neural network;

[0032] Step b6: Based on the constructed regression prediction network, construct the main loss function for model training; where the main loss function is denoted as Γ. main :

[0033]

[0034] Here, arg(·) refers to changing f b o x So that the function Γ reaches its minimum value, f j These are the model-predicted values ​​for the location of the exposed termite feature bounding box. It is the true value of the target location at the exposed feature bounding box of termites. The target true value Compared with the model's predicted value f j The matching loss between them, where N is the total number of termite indicator species that need to be predicted;

[0035] Step b7, construct the InfoNce loss function for the auxiliary self-supervised task; where the InfoNce loss function is denoted as Γ. ssl :

[0036]

[0037] f′ k =LOCEXT(IMGEXT(x′) k )), f″ k =LOCEXT(IMGEXT(x″) k ));

[0038] x′ k = aug(x k ), x″ k = aug(x′) k );

[0039] Where aug(·) is the image random augmentation function, x k It is the image to be detected, f′ k The image to be detected is x k An enhanced viewpoint image representation value, f″ k The image to be detected is x k Another enhanced perspective image representation value; τ is the temperature coefficient hyperparameter that needs to be set for model training, and B is the batch size for training;

[0040] Step b8: Based on the constructed main loss function and InfoNce loss function, the final loss function for model training is obtained; whereby the final loss function for model training is denoted as Γ:

[0041] Γ=ε1Γ main +ε2Γ ssl ;ε1+ε2=1;

[0042] Where ε1 is the first weight coefficient of the final loss function, and ε2 is the second weight coefficient of the final loss function;

[0043] Step b9: The target detection feature extraction model is trained using the backpropagation algorithm, and the converged target detection feature extraction model is used as the termite exposure feature detection model for the termite exposure feature detection.

[0044] In a further improvement, the process of the UAV acquiring sample images of the area to be detected includes the following steps in the aforementioned UAV-based intelligent detection method for termite exposure features:

[0045] Step S31: Command the drone to fly to the termite distribution area of ​​the area to be detected;

[0046] Step S32: The drone is instructed to fly at a preset altitude and preset speed over the termite distribution area at a constant speed.

[0047] Step S33: Adjust the shooting angle of the camera carried by the drone to collect multiple images of the termite distribution area under different lighting and different angles, and form a set of images to be detected using the collected multiple images.

[0048] Furthermore, in the intelligent detection method for exposed termite features based on drones, the process of adjusting the shooting angle of the camera carried by the drone in step S33 is as follows: rotating the shooting angle of the camera in the horizontal direction; and rotating the shooting angle of the camera in the vertical direction; wherein the rotation angle in the horizontal direction is in the range of 0 to 360°, and the rotation angle in the vertical direction is in the range of 0 to 180°.

[0049] In a further improvement, in the aforementioned intelligent detection method for exposed termite features based on drones, the preset exposed feature corresponding to the preset termite exposed feature image feature parameters is a termite nest location indicator.

[0050] Furthermore, in this invention, the intelligent detection method for exposed termite features based on drones also includes: performing a location positioning operation on the exposed termite feature based on the obtained detection results.

[0051] The technical solution adopted by the present invention to solve the second technical problem is: an ant nest location method, characterized in that it applies any of the above-mentioned intelligent detection methods for exposed termite features based on drones; wherein, the exposed termite features are termite nest location indicators.

[0052] In a further improvement, the termite nest location method in this invention includes: making a judgment based on the detection results of the intelligent detection method for exposed termite features: when exposed termite features are detected, it is determined that an ant nest exists in the actual location area corresponding to the exposed termite features; otherwise, it is determined that an ant nest does not exist in the actual location area corresponding to the exposed termite features; wherein, the exposed termite features are termite nest location indicators.

[0053] In a further improvement, the termite nest location method in this invention further includes: generating a termite nest location detection report for personnel to view based on the termite nest location results; or / and, generating a termite exposure feature inspection report for personnel to view based on the detection results of the intelligent termite exposure feature detection method; wherein the termite exposure feature inspection report includes the name of each termite exposure feature and the distribution location information of the corresponding termite exposure feature.

[0054] Compared with the prior art, the advantages of the present invention are as follows:

[0055] First, the intelligent detection method for exposed termite features based on drones of this invention pre-constructs a set of feature parameters for exposed termite images that characterize exposed termite features, and constructs a detection model for exposed termite features based on the set of feature parameters. Then, the image to be detected, which is obtained by the drone in the area to be detected and processed, is input into the detection model for exposed termite features. Based on the detection results of the detection model, it is determined whether the image to be detected contains exposed termite features, and further, based on the detection results of exposed termite features, it is determined whether there is termite activity in the area to be detected. This realizes intelligent detection of exposed termite features in the area to be detected, which is conducive to improving the efficiency and accuracy of the location of exposed termite features in the area to be detected, and further improving the accuracy of termite nest location.

[0056] Secondly, in constructing the termite exposure feature detection model, this invention uses a multi-layer Transformer module as a decoder for the target location information in the image representation, so as to realize end-to-end application. This allows for flexible adjustment of the model's own complexity to adapt to the real data distribution, effectively improving the generalization and robustness of the termite exposure feature detection model. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of the intelligent detection method for exposed termite features based on drones in an embodiment of the present invention. Detailed Implementation

[0058] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0059] This embodiment provides an intelligent detection method for exposed termite features based on unmanned aerial vehicles (UAVs). Exposed termite features include surface symbols of termite activity and termite nest location indicators. Surface symbols include things like termite trails and mud blankets, while termite nest location indicators include things like termite umbrellas, anthracite fungi, swarming holes, ventilation holes, excrement, and termite mounds. See also Figure 1 As shown, the intelligent detection method for exposed termite features based on drones in this embodiment includes the following steps S1 to S5:

[0060] Step S1: Pre-construct a set of image feature parameters representing the exposed features of termites; wherein, the set of image feature parameters representing the exposed features of termites includes at least one image feature parameter representing the exposed features of termites; in this embodiment, the set of image feature parameters representing the exposed features of termites includes image feature parameters representing swarming holes.

[0061] The pre-construction process of the termite exposure feature image feature parameter set is as follows: Step a1, pre-determine the target image feature parameter set to be detected when performing termite exposure feature detection on sample images acquired by UAV; and Step a2, construct the termite exposure feature image feature parameter set for identifying termite exposure features based on the determined target image feature parameter set.

[0062] Step S2: Construct a termite exposure feature detection model based on the feature parameter set of termite exposure feature images; that is, the termite exposure feature detection model constructed in this embodiment has at least the function of detecting swarming holes.

[0063] Step S3: Use a drone to acquire sample images of the area to be detected;

[0064] Step S4: Process the sample image to obtain the image to be detected;

[0065] Step S5: Input the image to be detected into the termite exposure feature detection model to obtain the termite exposure feature detection result for the image to be detected. For example, after detection based on the termite exposure feature detection model, if it is determined that image features representing swarming holes are detected, then it can be determined that termites exist in the area corresponding to the image to be detected.

[0066] Specifically, in this embodiment, the construction process of the termite exposure feature detection model in step S2 includes steps b1 to b9:

[0067] Step b1: Use a convolutional neural network to obtain a feature map of termite exposure features in the image to be detected; wherein, the feature map of termite exposure features is labeled as f. map_h ,:

[0068]

[0069] Where CNN(·) is the forward propagation function of the selected convolutional neural network CNN, x img Let H be the image to be detected, and let H be the image to be detected x. img The length of the image x to be detected is W, where W is the length of the image x to be detected. img The width of C1 represents the number of channels in the feature map output by the forward propagation function CNN(·).

[0070] Step b2: Perform dimensionality reduction on the termite exposure feature image to obtain a dimensionality-reduced feature map; wherein, the dimensionality-reduced feature map after the dimensionality reduction operation is labeled as f. map :

[0071] f map =conv 1×1 (f map_h ), f map ∈RC×H×W ;

[0072] Among them, f map Image feature map f representing termite indicator map_h The dimensionality-reduced feature map after dimensionality reduction operation, conv 1×1 (·) represents a convolution operation with a kernel size of 1 and a number of channels of C, where C represents the number of channels in the reduced feature map;

[0073] Step b3: Perform feature dimensionality reduction on the obtained dimensionality-reduced feature map to obtain the dimensionality-reduced feature vector, and construct the image feature extraction backbone model using this feature vector; wherein, the dimensionality-reduced feature vector is labeled f, and the image feature extraction backbone model is labeled IMGEXT(·):

[0074] f = σ(W1, (pad(f)) map ))), f∈R C×D ;

[0075] Where σ(·) represents the activation function of the feedforward neural network, W1 represents the learnable parameter matrix of the feedforward neural network, and pad(f map ) represents flattening a feature map of dimension H×W into a one-dimensional vector, where D is the dimension of the feature vector f;

[0076] Step b4: Based on the dimensionality-reduced feature vectors obtained from the features, construct a target detection feature extraction model; wherein, the target detection feature extraction model is denoted as LOCEXT(·):

[0077] f i+1 =Transformer i (f i ),f i ∈R C×D f1 = f, f i+1 ∈R C×D , i∈N+Λi≤L;

[0078] Among them, f i Let i represent the detection feature of the i-th object, and let Transformer be the transform. i (·) represents the forward propagation process of the i-th layer self-attention network Transformer, f L This represents the final target detection features;

[0079] Step b5: Based on the obtained final target detection features, construct a regression prediction network to predict the specific location of the termite exposure features within the image to be detected;

[0080] f b o x =max(0,f L W2+b2)W3+b3,fbox ∈R C×5 ;

[0081] Where max(0,f) L W2+b2) represents the relationship between the numerical values ​​0 and f. L The maximum value is taken from W2+b2, where W2, W3, b2, and b3 are learnable parameters in the feedforward neural network;

[0082] It should be noted that the so-called "specific location" in step b5 refers to the parameters of the rectangle that can just frame the exposed features of the termites, such as the coordinates of the upper left corner of the rectangle and its length and width.

[0083] Step b6: Based on the constructed regression prediction network, construct the main loss function for model training; where the main loss function is denoted as Γ. main :

[0084]

[0085] Here, arg(·) refers to changing f b o x So that the function Γ reaches its minimum value, f j These are the model-predicted values ​​for the location of the exposed termite feature bounding box. It is the true value of the target location at the exposed feature bounding box of termites. The target true value Compared with the model's predicted value f j The matching loss between them, where N is the total number of termite exposure feature species to be predicted;

[0086] Step b7, construct the InfoNce loss function for the auxiliary self-supervised task; where the InfoNce loss function is denoted as Γ. ssl :

[0087]

[0088] f′ k =LOCEXT(IMGEXT(x′) k )), f″ k =LOCEXT(IMGEXT(x″) k ));

[0089] x′ k = aug(x k ), x″ k = aug(x′) k );

[0090] Where aug(·) is the image random augmentation function, x k It is the image to be detected, f′ kThe image to be detected is x k An enhanced viewpoint image representation value, f″ k The image to be detected is x k Another enhanced view image representation value; τ is the temperature coefficient hyperparameter that needs to be set for model training (in this technical field, its full English name is Augmented-view image representation), and B is the training batch size;

[0091] Step b8: Based on the constructed main loss function and InfoNce loss function, the final loss function for model training is obtained; whereby the final loss function for model training is denoted as Γ:

[0092] Γ=ε1Γ main +ε2Γ ssl ;ε1+ε2=1;

[0093] Where ε1 is the first weight coefficient of the final loss function, and ε2 is the second weight coefficient of the final loss function;

[0094] Step b9: The target detection feature extraction model is trained using the backpropagation algorithm, and the converged target detection feature extraction model is used as the termite exposure feature detection model for the termite exposure feature detection.

[0095] In other words, after obtaining the image to be detected, it is input into the termite exposure feature detection model to obtain the final termite exposure feature detection result. That is, through detection based on the termite exposure feature detection model, it can be determined whether the image to be detected contains the selected fungi.

[0096] Regarding the acquisition of sample images of the area to be detected, in this embodiment, the process of the UAV acquiring sample images of the area to be detected includes the following steps S31 to S33:

[0097] Step S31: Command the drone to fly to the termite distribution area of ​​the area to be detected;

[0098] Step S32: The drone is instructed to fly at a preset altitude and preset speed over the termite distribution area at a constant speed.

[0099] Step S33: Adjust the shooting angle of the camera carried by the drone to collect multiple images of the termite distribution area under different lighting and angles, and form a set of images to be detected using these multiple images. The process of adjusting the shooting angle of the camera carried by the drone involves: rotating the camera's shooting angle horizontally; and rotating the camera's shooting angle vertically; wherein the horizontal rotation angle ranges from 0 to 360°, and the vertical rotation angle ranges from 0 to 180°.

[0100] Of course, depending on actual needs, the intelligent detection method for exposed termite features based on UAVs in this embodiment also includes: performing a location positioning operation on the exposed termite feature based on the obtained detection results.

[0101] This embodiment also provides a method for locating termite nests, which applies the aforementioned intelligent detection method for exposed termite features based on unmanned aerial vehicles (UAVs). Here, the exposed termite features are termite nest location indicators.

[0102] In the process of locating termite nests, the nest location method in this embodiment makes a judgment based on the detection results of an intelligent detection method for exposed termite features (i.e., nest indicators): when exposed termite features are detected, it is determined that a nest exists in the actual location area corresponding to the exposed termite features; otherwise, it is determined that a nest does not exist in the actual location area corresponding to the exposed termite features. Here, the exposed termite features are termite nest indicators.

[0103] To further improve the accuracy of termite nest location, the termite nest location method in this embodiment also adopts the following measures:

[0104] Using fungi as the detection target, the above-mentioned intelligent detection method for exposed termite features is executed again, and the termite nest location detected based on the fungi is used as the reference location closest to the actual location of the termite nest.

[0105] It should be noted that if *Termitomyces albuminosus* or *Stachys rubrum* is used as the detection target, and an ant nest is found in the actual location area corresponding to the detection target, it means that an active ant nest is located beneath the *Termitomyces albuminosus* or *Stachys rubrum*. Conversely, if *Carbohydratea charcoal* is used as the detection target, and an ant nest is found in the actual location area corresponding to the detection target, it means that a dead ant nest is located beneath the *Carbohydratea charcoal*.

[0106] To facilitate personnel's understanding of the location of termite nests, the termite nest location method in this embodiment also includes generating a termite nest location detection report for personnel to view based on the termite nest location results.

[0107] To facilitate personnel understanding the distribution location of exposed termite features, the termite nest location method in this embodiment also includes generating a termite exposure feature inspection report based on the detection results of an intelligent termite exposure feature detection method. This report includes the name of each exposed termite feature and its corresponding distribution location information. Specifically, the distribution location information here refers to the latitude and longitude coordinates of the termite exposure feature distribution. This latitude and longitude information can be obtained by personnel using a terminal for positioning or by a drone.

[0108] Although preferred embodiments of the present invention have been described in detail above, it should be clearly understood that various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting termite exposed features intelligently based on a UAV, characterized in that, The method comprises the following steps: Step S1, a termite exposed feature image feature parameter set representing termite exposed features is constructed in advance; wherein the termite exposed feature image feature parameter set comprises at least one image feature parameter representing the termite exposed features; Step S2, a termite exposed feature detection model is constructed based on the termite exposed feature image feature parameter set; Step S3, a sample image of a detection area is obtained by using a drone; Step S4, the sample image is processed to obtain a detection image; Step S5, the detection image is input into the termite exposed feature detection model to obtain a termite exposed feature detection result for the detection image; wherein: The pre-construction process of the termite exposed feature image feature parameter set is as follows: Step a1, a target image feature parameter set required for termite exposed feature detection on a sample image obtained by a drone is determined in advance; Step a2, a termite exposed feature image feature parameter set for identifying termite exposed features is constructed according to the determined target image feature parameter set; The construction process of the termite exposed feature detection model is as follows: Step b1, obtaining termite exposed feature image features for the to-be-detected image by using a convolutional neural network; wherein the termite exposed feature image features are marked as f map_h : f map_h = CNN(x img ), x img ∈ R 3×H×W ; wherein CNN(·) is a forward propagation function of a convolutional neural network CNN selected, x img is an image to be detected, H is a length of the image to be detected x img , W is a width of the image to be detected x img , and C1 represents a number of channels of a feature map output by the forward propagation function CNN(·). Step b2, dimension reduction operation is performed on the termite exposed feature image feature map to obtain a dimension-reduced feature map; wherein the dimension-reduced feature map after the dimension reduction operation is marked as f map : f map = conv 1×1 (f map_h ), f map ∈ R C×H×W ; wherein f map represents the exposed termite feature image feature map f map_h The dimensionality-reduced feature map after the dimensionality reduction operation, conv 1×1 (·) represents a convolution operation with a convolution kernel size of 1 and a channel number of C, and C represents the channel number of the dimensionality-reduced feature map; Step b3, a feature dimension reduction operation is performed on the obtained dimension-reduced feature map to obtain a dimension-reduced feature vector, and an image feature extraction backbone model is constructed based on the feature vector; wherein the dimension-reduced feature vector is marked as f, and the image feature extraction backbone model is marked as IMGEXT(·): f = σ(W1, (pad(f map ))), f ∈ R C×D ; where σ(·) denotes an activation function of the feedforward neural network, W1denotes a learnable parameter matrix of the feedforward neural network, pad(f map ) denotes flattening the feature map of dimension H x W into a one-dimensional vector, and D is the dimension of the feature vector f. Step b4, a target detection feature extraction model is constructed based on the obtained dimension-reduced feature vector; wherein the target detection feature extraction model is marked as LOCEXT(·): f i+1 = Transformer i (f i ), f i ∈ R C×D , fi = f, f i+1 ∈ R C×D , i ∈ N + Λi≤ L; wherein f i represents the i-th target detection feature, Transformer i (·) represents the forward propagation process of the i-th layer self-attention network Transformer, f L represents the final target detection feature; Step b5, a regression prediction network is constructed based on the obtained final target detection feature to predict the specific position of the termite exposed feature in the detection image; f box = max(0, f L W2 + b2) W3 + b3, f box ∈ R C×5 ; where max(0, f L W2+b2) represents the maximum of the value 0 and the value f L W2+b2, and W2, W3, b2, and b3 are learnable parameters in the feedforward neural network. Step b6, based on the constructed regression prediction network, constructing a main loss function for model training; wherein the main loss function is marked as Γ main : where arg(·) denotes the minimizer of f box such that the function Γ attains a minimum value, f j is the model prediction value of the exposed feature bounding box position of the termite, is the target real value of the exposed feature bounding box position of the termite, is the target real value is the matching loss between the model prediction value f j and the target real value, and N is the total number of termite indicator classes that need to be predicted. Step b7, constructing an InfoNce loss function of the auxiliary self-supervised task; wherein the InfoNce loss function is marked as Γ ssl : f k ' = LOCEXT(IMGEXT(x k )), f k ' = LOCEXT(IMGEXT(x k )) ; x' k = aug(x k ), x' k ' = aug(x' k ); where aug(·) is an image random augmentation function, x k is the image to be detected, f k is an augmented view image representation value of the image to be detected x k ; f k is another augmented view image representation value of the image to be detected x k ; τ is a temperature coefficient hyperparameter to be set for model training, and B is the batch size of training. Step b8, based on the constructed main loss function and InfoNce loss function, the final loss function for model training is obtained by processing; wherein the final loss function for model training is marked as Γ: Γ = ε1Γ main + ε2Γ ssl ; ε1+ ε2 = 1; Wherein ε1 is the first weight coefficient of the final loss function, and ε2 is the second weight coefficient of the final loss function; Step b9, the target detection feature extraction model is trained by using a back propagation algorithm, and the converged target detection feature extraction model is used as the termite exposed feature detection model for termite exposed feature detection.

2. The unmanned aerial vehicle based termite external feature intelligent detection method according to claim 1, characterized in that, The process of obtaining a sample image of a detection area by using a drone comprises the following steps: Step S31, the drone is flown to a termite distribution area of the detection area; Step S32, the drone is flown at a preset flight height and a preset flight speed through the termite distribution area; Step S33, the shooting angle of the camera carried by the drone is adjusted to respectively collect multiple images of the termite distribution area under different illuminations and different angles, and the collected multiple images form a detection image set.

3. The unmanned aerial vehicle based termite external feature intelligent detection method according to claim 2, characterized in that, In step S33, the process of adjusting the shooting angle of the camera carried by the UAV is: rotating the shooting angle of the camera in the horizontal direction; and rotating the shooting angle of the camera in the vertical direction; wherein the rotation angle range in the horizontal direction is 0-360°, and the rotation angle range in the vertical direction is 0-180°.

4. The unmanned aerial vehicle based termite external feature intelligent detection method according to any one of claims 1-3, characterized in that, The preset exposed feature corresponding to the termite exposed feature image feature parameter is a termite nest site indicator.

5. The UAV-based termite external feature intelligent detection method according to claim 4, characterized in that, Further comprising: Based on the obtained termite exposed feature detection result, a position positioning operation is performed on the termite exposed feature.

6. A method of nest positioning, characterized in that The application has the unmanned aerial vehicle-based termite exposed feature intelligent detection method according to any one of claims 1-5; wherein the termite exposed feature is a termite nest site indicator.

7. The ant nest location method of claim 6, wherein, Further comprising: According to the detection result of the termite exposed feature intelligent detection method, a judgment process is performed: when the termite exposed feature is detected, it is determined that the actual position area corresponding to the termite exposed feature has a termite nest; otherwise, it is determined that the actual position area corresponding to the termite exposed feature does not have a termite nest; wherein the termite exposed feature is a termite nest site indicator.

8. The ant nest location method of claim 7, wherein, Further comprising: Based on the termite nest positioning result, a termite nest positioning detection report for personnel to view is generated; or / and, based on the detection result of the termite exposed feature intelligent detection method, a termite exposed feature inspection report for personnel to view is generated; wherein the termite exposed feature inspection report includes the names of various termite exposed features and the distribution position information of the corresponding termite exposed features.

Images