Tick monitoring and classifying system based on deep learning

Abstract

The invention relates to the technical field of monitoring of vector organisms, and discloses a tick monitoring and classifying system based on deep learning, and the system obtains continuous image data containing ticks in a monitoring region through an image collection unit, carries out the background separation and candidate region extraction of the image data through a preprocessing unit, and carries out the classification of ticks. Filtering and generating candidate target images in combination with morphological constraints; performing multi-scale feature extraction on the candidate target image through a target detection unit to realize automatic positioning of the tick target; the classification and recognition unit is further used for extracting morphological characteristics such as the body shape proportion, the foot structure and the back plate texture of the ticks, and classifying and recognizing the types of the ticks; meanwhile, the position change of the ticks in continuous time frames is analyzed through a behavior analysis unit, and movement behavior parameters of the ticks are obtained; and finally, the evaluation unit synthesizes the tick quantity, variety and behavior information to generate tick activity indexes of the monitoring area.

Description

Technical Field

[0001] This invention relates to the field of vector-borne disease monitoring technology, and more specifically, to a tick monitoring and classification system based on deep learning. Background Technology

[0002] Ticks are important blood-sucking arthropods widely distributed in grasslands, woodlands, shrublands, and livestock farming areas. During their blood-sucking process, they can transmit various pathogens and are important vectors for many zoonotic diseases. With changes in the ecological environment, the expansion of wildlife ranges, and the increased frequency of human outdoor activities, the risk of tick bites and related diseases is on the rise, posing a continuous threat to public health and livestock production. Therefore, long-term and effective monitoring of tick distribution, activity status, and potential risks is of significant practical importance.

[0003] Current tick monitoring and survey methods mainly rely on manual patrols and laboratory identification and analysis. These methods typically require professionals to periodically deploy sampling devices in the field or breeding environments, collect samples, and perform manual observation and classification. This approach suffers from high labor intensity, long monitoring cycles, and limited spatial coverage, making it difficult to meet the practical needs of large-scale, continuous monitoring. Furthermore, the manual identification process heavily relies on the professional experience of the operators, and inconsistencies in interpretation results among different personnel can affect the stability and comparability of monitoring data.

[0004] In recent years, with the development of image acquisition equipment and computer vision technology, image-based vector identification methods have gradually attracted attention. Some technologies attempt to acquire images of target organisms using camera equipment and then use image processing or machine learning methods to identify and count the targets. However, ticks are small in size, their color is similar to the environmental background, and they are often found in complex backgrounds such as vegetation and soil. Traditional methods based on single-scale features or handcrafted features are easily affected by changes in lighting, background interference, and target occlusion in real-world environments, making it difficult to guarantee detection stability and identification accuracy.

[0005] Furthermore, existing technologies primarily focus on identifying or counting targets in single static images, typically only determining the presence or approximate number of ticks in a monitored area. They lack continuous tracking and behavioral analysis of ticks over time. Ticks may exhibit different activity states at different stages, such as lurking, actively crawling, or attaching, and these behavioral characteristics are closely related to the risk of biting. Relying solely on static image results makes it difficult to accurately reflect the actual activity level and potential risk of ticks.

[0006] Meanwhile, existing deep learning-based object detection or classification schemes often directly apply general object detection models to vector-borne disease identification tasks, lacking specific designs for tick morphological characteristics. For example, tick species exhibit subtle differences in body proportions, leg structure, and dorsal plate texture. Without targeted feature modeling, general models are prone to confusing closely related species. Furthermore, current technologies typically treat detection, classification, and risk assessment as independent processes, lacking a unified data flow and logical connection, making it difficult to form a complete technical chain from raw monitoring data to regional risk assessment results.

[0007] Therefore, existing technologies in the field of tick monitoring still have the following shortcomings: it is difficult to stably detect tiny tick targets in complex natural environments, it is difficult to distinguish tick species in detail, there is a lack of systematic analysis of the continuous movement trajectory and behavioral status of ticks, and there is a lack of regional-level evaluation mechanisms that effectively integrate detection results, classification results and behavioral information.

[0008] Therefore, there is an urgent need for a deep learning-based tick monitoring and classification system to solve these problems. Summary of the Invention

[0009] The purpose of this invention is to solve the technical problems mentioned in the background section and to provide a tick monitoring and classification system based on deep learning.

[0010] The objective of this invention is achieved through the following technical solution: A deep learning-based tick monitoring and classification system includes: The data acquisition unit is used to acquire continuous image data containing ticks within the monitoring area; The preprocessing unit is used to perform background separation, image cropping, and scale normalization on the continuous image data to generate candidate target images; The target detection unit is used to extract convolutional features from the candidate target image, locate the tick target based on the multi-scale feature map, and output the bounding box position of the tick target. A classification and recognition unit is used to encode the morphological features of the tick image within the bounding box and classify the tick species based on the morphological features; The behavior analysis unit is used to analyze the spatial position changes of the same tick in consecutive time frames and generate tick movement behavior parameters; The evaluation unit is used to evaluate the tick activity in the monitoring area and generate tick activity indicators based on the tick classification results and the movement behavior parameters.

[0011] As a preferred technical solution of the present invention, the data acquisition unit includes a macro imaging device, which has imaging resolution capability that can distinguish the structure of tick feet and the texture features of the dorsal plate.

[0012] As a preferred technical solution of the present invention, the preprocessing unit performs background subtraction processing on consecutive image frames to remove static vegetation and ground background information, and retains only the suspected biological movement area as the candidate target image.

[0013] As a preferred embodiment of the present invention, the target detection unit includes: The first feature extraction layer is used to extract shallow texture features; The second feature extraction layer is used to extract high-level semantic features; And a feature fusion layer for fusing features at different scales; The fused features output by the feature fusion layer are used to generate the bounding box location of the tick target.

[0014] As a preferred technical solution of the present invention, the morphological features of ticks extracted by the classification and identification unit include at least the body length-to-width ratio feature, the number distribution feature of the legs, and the directional feature of the dorsal plate texture, and the tick species are identified based on the morphological features.

[0015] As a preferred technical solution of the present invention, the classification and identification unit maps the morphological features of the ticks to a unified feature space and distinguishes different tick species by comparing feature similarity.

[0016] As a preferred technical solution of the present invention, the behavior analysis unit obtains the tick's movement speed information, movement direction change information, and dwell time information based on the changes in the center position of the tick's bounding box in continuous time frames.

[0017] As a preferred technical solution of the present invention, the behavior analysis unit classifies tick behavior into stationary behavior, crawling behavior, or attachment behavior based on the movement speed information and dwell time information.

[0018] As a preferred technical solution of the present invention, when generating tick activity indicators, the evaluation unit comprehensively considers the distribution of tick numbers in the monitoring area, the risk weight corresponding to tick species, and the behavioral weight corresponding to tick behavior type.

[0019] As a preferred technical solution of the present invention, the tick activity index is generated by comprehensively considering different tick species and different behavioral types according to a preset weighting rule.

[0020] Compared with the prior art, the present invention has the following beneficial effects: 1. The tick monitoring and classification system based on deep learning provided by this invention combines continuous image acquisition, background difference preprocessing, candidate region morphological constraints and multi-scale target detection to achieve automatic localization and stable detection of tick targets in complex natural environments. This technical solution can effectively suppress the interference of vegetation texture, surface noise and light changes on the detection results, improve the detection reliability of tiny tick targets in complex backgrounds such as grasslands and forest edges, reduce the dependence on manual patrol and screening, and is suitable for long-term and continuous tick monitoring application scenarios.

[0021] 2. Building upon tick detection, this invention further introduces refined modeling of tick morphological characteristics. By jointly encoding features such as body length-to-width ratio, leg number distribution, and dorsal plate texture directionality, it achieves precise differentiation between different tick species. Compared to identification methods relying solely on general convolutional features, this technical solution effectively reduces misclassification among closely related species, improves the stability and consistency of tick classification results, and provides more reliable basic data support for subsequent risk analysis.

[0022] 3. This invention analyzes the spatial location changes of ticks across continuous time frames to construct tick movement trajectories and extract behavioral parameters such as movement speed and dwell time. Based on this, the behavioral status of ticks is determined, and the behavioral information is integrated with tick quantity and species information to generate regional-scale tick activity indicators. This technical solution overcomes the limitation of traditional static identification methods that can only reflect "the presence or absence of ticks," achieving a quantitative assessment of tick activity levels. This facilitates the identification and hierarchical management of high-risk areas, providing a more comprehensive and intuitive technical basis for tick control and public health risk decision-making. Attached Figure Description

[0023] Figure 1 This is a system block diagram of the present invention; Figure 2 This is a diagram showing the composition of the target detection unit of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with embodiments and appendices. Figures 1-2 The present invention will be further described in detail below. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0025] Example 1: In one specific embodiment of the present invention, a deep learning-based tick monitoring and classification system is deployed in grasslands, forest edges, or livestock activity areas. A data acquisition unit continuously acquires images of the monitored area to obtain continuous image data containing tick activity information. The data acquisition unit preferably employs an imaging device with macro imaging capabilities, whose imaging resolution can clearly distinguish the tick's foot structure and dorsal plate texture features, thus providing a reliable image foundation for subsequent target detection and classification.

[0026] Let the first The original images acquired at each time point are ,in This indicates the time frame number. To eliminate interference from static vegetation, surface texture, and slow changes in lighting on tick detection, the preprocessing unit first performs background subtraction processing on adjacent time frame images to obtain a difference response map: ; in, Indicates the first Differential response map of frames, and They represent the first Frame and the The original image of the frame. This indicates the operation of taking the absolute value pixel by pixel.

[0027] Obtaining the differential response map Then, threshold segmentation is performed on it to generate a binary motion mask, specifically represented as follows: ; in, This indicates the binary motion mask at pixel coordinates. The value at; This indicates the difference response map in pixel coordinates. Pixel value at; This is the differential threshold used to distinguish significant motion regions from background noise; Image pixel coordinates in obtaining a binary motion mask Then, the mask is processed using the connected component extraction operator to obtain a set of candidate regions: ; in, Indicates the connected component extraction operator; Indicates the first The first frame The set of pixels corresponding to each connected component; Indicates the first The number of candidate connected components in a frame.

[0028] To further eliminate non-tick target areas caused by wind-blown grass, shadow disturbance, etc., each candidate area was analyzed. Apply shape constraint filtering.

[0029] First, calculate the area of ​​the candidate region: ; in, Indicates candidate region The number of pixels it contains Let represent the cardinality of the set. Then, calculate the height and width of the minimum bounding rectangle of the candidate region, denoted as [x, y]. and And thus the aspect ratio of the candidate region is obtained: ; in, This indicates the morphological and proportional characteristics of the candidate region.

[0030] A candidate region is retained for subsequent processing only if it meets the area threshold and aspect ratio threshold constraints. ; in, These are the lower and upper thresholds for the candidate region area, respectively. These represent the lower and upper aspect ratio thresholds for the candidate regions, respectively. Cropping and scale normalization are performed on candidate regions that meet the morphological constraints to obtain candidate target image patches. And input it into the target detection unit.

[0031] The target detection unit employs a multi-scale target detection network that includes a shallow texture feature extraction layer and a high-level semantic feature extraction layer, and enhances the detection capability of tiny tick targets through a feature fusion structure.

[0032] For each candidate target image patch The target detection unit outputs the corresponding bounding box location of the tick target: ; in, They represent the first The first frame The x and y coordinates of the center point of the bounding box of the tick target; These represent the width and height of the bounding box, respectively.

[0033] After obtaining the spatial localization results of the tick target, the classification and recognition unit encodes the morphological features of the tick image within the bounding box. First, it directly calculates the aspect ratio of the tick body based on the bounding box size: ; in, Indicates the first The first frame The aspect ratio characteristics of individual tick targets.

[0034] Secondly, to characterize the directional texture of the tick's dorsal plate, gradient direction statistics were performed on the tick image within the bounding box, and the gradient direction was set at pixel coordinates. The gradient magnitude at that point is The gradient direction is Then the first The texture energy within a directional interval can be represented as: ; in, Indicates the first Texture energy within a directional range; Indicates the first One gradient direction interval; This is an indicator function that takes the value 1 when the condition is true and 0 otherwise.

[0035] Furthermore, by performing local response analysis on the foot region of tick images, the distribution characteristics of foot numbers were obtained. Let the... Key areas of the foot are The foot characteristic response is then expressed as: ; in, Indicates the first The response intensity of each foot region; This indicates an enhanced response in the foot area; Indicates the first Key areas of the foot.

[0036] By combining the above-mentioned body length-to-width ratio characteristics, dorsal plate texture directionality characteristics, and foot number distribution characteristics, a unified representation of tick morphological characteristics is formed: ; in, The backplate texture directional feature vector. This is the characteristic vector of the foot quantity distribution.

[0037] The classification and recognition unit represents morphological features. Mapping to a unified feature space yields the embedded feature vectors: ; in, Represents the embedded feature vector. This represents the feature mapping function. The system pre-stores the category prototype vectors corresponding to each tick species. The tick species were identified by comparing feature similarity. , ; in, Indicates the first The first frame The tick target and the first Similarity between tick-like organisms; Indicates the first Tick-like prototype feature vectors; This indicates the final tick species label obtained.

[0038] After detecting and classifying tick targets, the behavior analysis unit correlates the spatial locations of the same tick across consecutive time frames to form tick movement trajectories. Let the... The trajectory in the first The center position in the frame is: ; in, Coordinates of the center of the bounding box from the corresponding tick target.

[0039] Calculate the instantaneous movement speed of the tick based on the positional changes in adjacent time frames: ; in, Indicates the first The first frame The instantaneous speed of movement of the trajectory; Indicates Euclidean distance; This indicates the time interval between adjacent frames.

[0040] The average movement speed of ticks was obtained by statistically analyzing the velocity sequence. And based on the speed being below the stationary threshold in multiple consecutive frames. The duration of the tick's stay was used to calculate the tick's residence time parameter. Based on the parameters of average movement speed and dwell time, tick behavior is discriminated and classified into stationary behavior, crawling behavior, or attachment behavior, thereby obtaining the behavioral state corresponding to each trajectory.

[0041] Finally, the assessment unit generates tick activity indices by integrating tick population, tick species, and tick behavioral parameters at the regional scale. Let the monitoring area be... The number of tick tracks detected within the statistical time window was The tick population density is: ; in, This indicates the density of ticks within the monitored area.

[0042] Let the first The species weights corresponding to the tick tracks are: The behavior weight corresponding to its behavior state is The tick activity index in the monitoring area can then be expressed as: ; in, Indicators of tick activity in the monitored area; Used to characterize the differences in the contribution of different tick species to regional risk assessment; Used to characterize the degree to which different tick behavioral states affect activity levels.

[0043] Through the above continuous processing flow, the present invention realizes a complete chain processing process from original image acquisition, candidate region generation, tick target detection, morphological feature extraction, species classification, behavioral trajectory analysis to regional activity index evaluation. It can realize automatic monitoring and classification of ticks in complex natural environments and provide reliable technical support for tick control and public health risk management.

[0044] Example 2: In this example, a deep learning-based tick monitoring and classification system is applied to a tick monitoring scenario in the boundary area between farmland, grassland, and forest edge. The system uses a mobile monitoring device to conduct inspections in the fields to dynamically acquire information on the distribution and activity status of ticks.

[0045] The mobile monitoring equipment includes a movable carrier, an image acquisition device, a data processing unit, and a wireless communication module. The movable carrier can be a manually propelled small inspection vehicle, a tracked agricultural robot, or a mounted mobile platform, capable of slowly moving along field ridges, between crop rows, or through livestock passages. The image acquisition device is installed below or at the front of the movable carrier, maintained at a preset height above the ground surface to perform close-range imaging of surface vegetation and soil. Preferably, the image acquisition device has macro imaging capabilities, enabling it to clearly capture the body structure and surface texture features of ticks.

[0046] During the inspection, the mobile monitoring equipment moves through the fields at a constant or intermittent speed. The image acquisition device continuously acquires surface images at preset time intervals and transmits the acquired continuous image data to the data processing unit in real time. The data processing unit performs background separation and candidate region generation operations on the continuous image data, automatically removing static vegetation and soil background, and retaining only areas of suspected biological activity as candidate target images.

[0047] Subsequently, the data processing unit performs multi-scale target detection processing on the candidate target images, automatically locating tick targets in complex backgrounds and outputting the spatial position of the ticks in the images. For the detected tick targets, the system further extracts their body proportions, foot structure distribution, and dorsal plate texture features, and classifies and identifies the types of ticks based on these morphological features, thereby distinguishing different tick species and developmental stages.

[0048] Based on the above embodiments, the classification and identification unit of the present invention is not only used to identify tick species, but also to identify the developmental stages of ticks. Ticks exhibit significant morphological differences at different developmental stages, particularly in the number of legs. Larval ticks have six legs, while nymphal or adult ticks have eight legs. Therefore, the distribution of the number of legs can serve as an important criterion for distinguishing tick developmental stages.

[0049] In this embodiment, after the classification and recognition unit completes the target localization of the tick, it performs foot region enhancement processing on the tick image within the bounding box. By detecting the slender structures near the tick outline, it extracts the local response region corresponding to the tick's foot. Specifically, the system constructs a set of candidate foot regions in the tick image and statistically analyzes the response intensity and spatial distribution of each candidate region to obtain the tick foot quantity distribution characteristics.

[0050] After obtaining the distribution characteristics of the number of legs, the classification and identification unit makes a preliminary determination of the tick's developmental stage based on the number of valid leg areas detected. When six legs are detected, the tick is classified as a larva; when eight legs are detected, the tick is classified as an adult or nymph. To improve the stability of the determination, the leg number determination results for the same tick target are checked for consistency across consecutive time frames. When the leg number determination results are consistent across multiple consecutive frames, the developmental stage determination result for the tick is confirmed.

[0051] In some implementations, the classification and identification unit also performs joint analysis of the foot number distribution characteristics, body length-to-width ratio characteristics, and dorsal plate texture directionality characteristics. Since ticks exhibit differences in body proportions and dorsal plate texture clarity at different developmental stages, the multi-feature joint discrimination method can further reduce misjudgments caused by occlusion, posture changes, or image blurring, thereby improving the reliability of tick developmental stage discrimination.

[0052] Through the above methods, the classification and identification unit of the present invention can automatically identify the developmental stage of ticks while completing the classification of tick species. This enables the system to not only identify "which type of tick" but also "what developmental stage it is in," providing more refined data support for tick activity assessment and risk analysis.

[0053] During mobile inspections, as the equipment moves continuously through the fields, the same tick may be repeatedly captured in multiple consecutive image frames. The system automatically constructs the tick's movement trajectory by correlating the spatial changes of the tick's position in these consecutive images, and analyzes the tick's movement speed and dwell time accordingly. When a tick remains in a certain area for an extended period or exhibits slow movement, the system classifies its behavior as stationary or attaching; when the tick shows continuous movement in the image, it is classified as crawling.

[0054] After completing tick detection, classification, and behavioral analysis, the system comprehensively statistically analyzes the number, species distribution, and behavioral status of ticks in the current inspection area, generating tick activity indicators for the corresponding area. These tick activity indicators can reflect the relative level of tick activity in different areas of the field and are transmitted in real time to a remote monitoring terminal or management platform via a wireless communication module, enabling managers to focus on high-risk areas or take targeted control measures.

[0055] As can be seen from the above embodiments, the tick monitoring and classification system provided by the present invention can adapt to the complex and ever-changing actual environment in the field, and realize the automatic detection, species identification, behavior analysis and regional activity assessment of ticks under mobile inspection conditions, which has good practicality and promotion value.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A tick monitoring and classification system based on deep learning, characterized in that, include: The data acquisition unit is used to acquire continuous image data containing ticks within the monitoring area; The preprocessing unit is used to perform background separation, image cropping, and scale normalization on the continuous image data to generate candidate target images; The target detection unit is used to extract convolutional features from the candidate target image, locate the tick target based on the multi-scale feature map, and output the bounding box position of the tick target. A classification and recognition unit is used to encode the morphological features of the tick image within the bounding box and classify the tick species based on the morphological features; The behavior analysis unit is used to analyze the spatial position changes of the same tick in consecutive time frames and generate tick movement behavior parameters; The evaluation unit is used to evaluate the tick activity in the monitoring area and generate tick activity indicators based on the tick classification results and the movement behavior parameters.

2. The tick monitoring and classification system based on deep learning according to claim 1, characterized in that, The data acquisition unit includes a macro imaging device, which has imaging resolution capabilities that can distinguish the structure of the tick's foot and the texture features of its dorsal plate.

3. The tick monitoring and classification system based on deep learning according to claim 1, characterized in that, The preprocessing unit performs background subtraction on consecutive image frames to remove static vegetation and ground background information, retaining only suspected biological movement areas as candidate target images.

4. The tick monitoring and classification system based on deep learning according to claim 1, characterized in that, The target detection unit includes: The first feature extraction layer is used to extract shallow texture features; The second feature extraction layer is used to extract high-level semantic features; And a feature fusion layer for fusing features at different scales; The fused features output by the feature fusion layer are used to generate the bounding box location of the tick target.

5. The tick monitoring and classification system based on deep learning according to claim 1, characterized in that, The morphological features of ticks extracted by the classification and identification unit include at least the body length-to-width ratio, the number and distribution of legs, and the directional features of the dorsal plate texture. Based on the morphological features, the tick species and tick developmental stage are determined.

6. The tick monitoring and classification system based on deep learning according to claim 1, characterized in that, The classification and identification unit maps the morphological features of the ticks to a unified feature space and distinguishes different tick species by comparing feature similarity.

7. The tick monitoring and classification system based on deep learning according to claim 1, characterized in that, The behavior analysis unit obtains information on the tick's movement speed, movement direction, and dwell time based on the changes in the center position of the tick's bounding box in continuous time frames.

8. The tick monitoring and classification system based on deep learning according to claim 7, characterized in that, The behavior analysis unit classifies tick behavior into stationary behavior, crawling behavior, or attachment behavior based on the movement speed information and dwell time information.

9. The tick monitoring and classification system based on deep learning according to claim 1, characterized in that, When generating tick activity indicators, the evaluation unit comprehensively considers the distribution of tick numbers in the monitoring area, the risk weights corresponding to tick species, and the behavioral weights corresponding to tick behavior types.

10. The deep learning-based tick monitoring and classification system according to claim 9, characterized in that, The tick activity index is generated by combining different tick species and different behavioral types according to a preset weighting rule.

Images