Insect identification camera and insect identification method using same

Abstract

The invention discloses an insect identification camera, and the camera comprises an image collection module which comprises a macro lens with the optical focal length of 3-5cm and a light supplement lamp with the color temperature in a range of 3000K-6000K and capable of being adjusted in a stepless manner; the storage module is used for storing an insect appearance feature library and an insect wound corresponding library; the processing module is used for operating a recognition algorithm, extracting image features and comparing the image features with the database; the display module is used for displaying the image and the recognition result; and the power supply module supplies power to each module. The insect identification camera can accurately and rapidly identify an insect through a high-quality microspur image in combination with the shape of the insect and the shape of a wound.

Description

Technical Field

[0001] This invention relates to the field of insect identification, and more specifically, to an insect identification camera and an insect identification method using the same. Background Technology

[0002] Insect bites are a common problem encountered by humans during outdoor activities, agricultural production, and daily life. The severity of injury and treatment methods vary greatly depending on the type of insect bite. For example, a bee sting requires immediate removal of the stinger and alkaline treatment, while a tick bite requires complete removal using professional techniques to prevent mouthparts from remaining. Therefore, quickly and accurately identifying the causative insect type is crucial for effective treatment.

[0003] Currently, people mainly rely on personal experience to identify harmful insects. However, ordinary people find it difficult to accurately distinguish between insects that look similar, especially for uncommon insects, where the misjudgment rate is extremely high and it is easy to delay treatment. Summary of the Invention

[0004] The purpose of this invention is to provide an insect identification camera and an insect identification method using the same. The insect identification camera can accurately and quickly identify insects by combining high-quality macro images with the shape of the insect's appearance and the shape of its wound.

[0005] To achieve the above objectives, the present invention provides an insect identification camera, comprising: The image acquisition module includes a macro lens with an optical focal length of 3cm to 5cm and a fill light with a color temperature that can be steplessly adjusted in the range of 3000K to 6000K. The storage module is used to store a database of insect morphological features and a database corresponding to insect wounds; The processing module is used to run the recognition algorithm, extract image features, and compare them with the database; The display module is used to display images and recognition results; The power supply module provides power to each module.

[0006] Preferably, the supplementary light includes multi-color temperature LEDs, and the brightness of each LED is independently controlled by pulse width modulation technology to achieve color temperature adjustment; the processing module is configured to automatically call up a supplementary light scheme of about 4500K in wound recognition mode and a supplementary light scheme of about 5500K in shape recognition mode.

[0007] Preferably, the storage module adopts an embedded multimedia card, and the insect morphological feature library and the insect wound correspondence library are stored in a structured data format. The records include standard images, text descriptions and feature vectors. The feature vectors include at least one of Hu moment-based shape descriptors, color histogram data, local binary pattern texture features and deep features extracted by a pre-trained convolutional neural network.

[0008] Preferably, the processing module uses an embedded system-on-a-chip with a neural network processing unit and is equipped with a lightweight embedded operating system.

[0009] The present invention also provides an insect identification method using the insect identification camera, comprising the following steps: Images of insects or wounds are acquired using the image acquisition module; The processing module extracts the shape features or wound features from the image; The extracted features are compared with a pre-stored database of insect morphological features or a database of insect wound correspondences. The display module outputs the matching insect type and treatment suggestions.

[0010] Preferably, before feature extraction, the image is preprocessed, and the preprocessing includes at least one of the following operations: noise reduction using nonlocal mean or bilateral filtering algorithms, enhancement using contrast-limited adaptive histogram equalization, and segmentation of the wound image using region growing or graph cut algorithms.

[0011] Preferably, in shape recognition mode, the extracted features include: Shape features described by calculating Hu invariant moments; Color histogram characteristics in the HSV color space; Texture features extracted using local binary patterns or Gabor filters; Deep features extracted using a lightweight convolutional neural network.

[0012] Preferably, in wound recognition mode, the extracted features include: At least one morphological feature among the area, perimeter, equivalent diameter, and compactness of the segmented inflamed region is calculated. Shape features described by Hu invariant moments; The swollen area has at least one color feature in the mean R value of the RGB channel and the mean S value in the HSV space; The pixel analysis features in the central region are used to determine whether there are puncture holes or mouthpart remnants.

[0013] Preferably, comparing the extracted features with the database includes: Calculate the Euclidean distance or cosine similarity between the feature vector of the image to be identified and the feature vector of the template in the database, and determine the insect type based on the K-nearest neighbor algorithm or support vector machine; and / or The image to be identified or its feature map is input into a lightweight neural network classifier to directly obtain the probability distribution of insect species.

[0014] Preferably, when insect images and wound images are input simultaneously, the processing module initiates the shape recognition and wound recognition processes in parallel, and combines the recognition results and confidence levels of the two processes; if the results are consistent, the overall confidence level is increased; if the results are inconsistent, a decision is made based on a preset rule base, including prioritizing the result with higher confidence or prompting the user of information conflict.

[0015] According to the above technical solution, this invention identifies insects by photographing their appearance and the wounds inflicted by insect bites, and then comparing the photographed images with appearance feature databases and wound feature databases. Two identification modes are used to identify insect species separately, and cross-validation can further improve the accuracy of identification.

[0016] The macro lens and adjustable fill light ensure high-quality images with clear details, while the image sensor and processing module guarantee image quality and processing capabilities. The built-in database and processing module allow the insect recognition camera to identify insects without an internet connection, enabling users to place it in remote locations and use it for simple self-rescue when bitten by insects.

[0017] The insect recognition camera also includes a display module that shows the recognition results along with processing suggestions. Users can determine the processing method based on the recognition results or refer to the processing suggestions provided by the insect recognition camera.

[0018] Therefore, this portable device can accurately and quickly identify insects even in environments without a network by using high-quality macro images combined with the insect's shape and the shape of its wounds.

[0019] Other features and advantages of the present invention will be described in detail in the following detailed description section. Detailed Implementation

[0020] The following provides a detailed description of specific embodiments of the present invention. It should be understood that the specific embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the scope of the invention.

[0021] An insect identification camera, comprising: The image acquisition module includes a macro lens, a fill light, and an image sensor; The storage module includes a built-in database of insect morphological features and a database corresponding to insect wounds. The processing module is used to extract image features and compare them with the database; The display module is used to output the recognition results and processing suggestions. The power module supports offline use.

[0022] By implementing the above technical solution, this insect recognition camera captures images of the insect's appearance and the bite wound, then compares the captured images with appearance feature databases and wound feature databases to identify the insect. Two recognition modes are used to identify the insect species separately, and cross-validation can further improve the recognition accuracy.

[0023] The macro lens and adjustable fill light ensure high-quality images with clear details, while the image sensor and processing module guarantee image quality and processing capabilities. The built-in database and processing module allow the insect recognition camera to identify insects without an internet connection, enabling users to place it in remote locations and use it for simple self-rescue when bitten by insects.

[0024] The insect recognition camera also includes a display module that shows the recognition results along with processing suggestions. Users can determine the processing method based on the recognition results or refer to the processing suggestions provided by the insect recognition camera.

[0025] Therefore, even in a network-free environment, this insect recognition camera can accurately and quickly identify insects by combining high-quality macro images with their shape and the shape of their wounds.

[0026] In this embodiment, preferably, the focal length of the macro lens is no greater than 5cm.

[0027] In this embodiment, preferably, the color temperature of the fill light is adjustable to adapt to different shooting scenarios.

[0028] In this embodiment, preferably, the resolution of the image sensor is not less than 12 million pixels.

[0029] In this embodiment, preferably, the processing module supports shape recognition mode, which includes extracting the insect's body shape, wing veins, and antennae features.

[0030] In this embodiment, preferably, the processing module also supports a wound recognition mode, which includes extracting the shape of the swollen area, the puncture point, and the redness features.

[0031] In this embodiment, preferably, the insect morphological feature library and the insect wound correspondence library in the storage module can be accessed offline.

[0032] In this embodiment, preferably, the display module outputs treatment suggestions for the insect bite while outputting the recognition result.

[0033] In one embodiment, the macro lens can be a fixed-focus or autofocus macro lens with an optical focal length designed to be 3cm to 5cm, to ensure that high-resolution, low-distortion images can still be obtained at extremely close focusing distances. For example, the lens can clearly capture the limb details of a chigger mite that is only 2-3mm in length.

[0034] The supplementary light consists of a group of multi-color temperature LEDs (such as cool white LEDs and warm white LEDs), and the brightness of each LED is independently controlled by pulse width modulation (PWM) technology to achieve stepless adjustment of color temperature within the range of 3000K to 6000K.

[0035] Preferably, in the "wound recognition mode", the system can automatically call a supplementary lighting scheme with a color temperature of about 4500K and soft light to restore the skin redness, blisters and other features with the best contrast; in the "shape recognition mode", a near-sunlight scheme with a color temperature of about 5500K can be called to realistically reproduce the color and texture of insects.

[0036] Image sensor: A CMOS image sensor with at least 12 million effective pixels and good low-light signal-to-noise ratio performance is preferred. The sensor is precisely aligned with the macro lens to ensure a consistent imaging plane.

[0037] In one embodiment, an embedded multimedia card (eMMC) is used as the storage medium, with a capacity of not less than 64GB. The module internally stores a database of insect morphological characteristics, a database of insect wound correspondences, and system programs.

[0038] The insect morphological feature database is stored in a structured data format. Each record corresponds to one insect and includes its standard image and feature vectors annotated by experts. Feature vectors may include, but are not limited to: shape descriptors based on Hu moments, histogram data representing color distribution, texture features extracted through Local Binary Patterns (LBP), and depth features extracted through pre-trained convolutional neural networks (CNNs).

[0039] The insect wound database is also stored in a structured data format. Each record is associated with one or more insects and includes an image of its typical wound, a text description (such as "linear arrangement" or "with a black dot in the center"), and a quantized feature vector. The feature vector may include: the approximate diameter of the swollen area, the ratio of its perimeter to its area, the presence of a bright pixel in the center (representing a puncture or mouthpart remnant), and the average color saturation of the swollen area.

[0040] The processing module exists as the core computing unit. In one embodiment, an embedded system-on-chip (SoC) with a neural network processing unit (NPU), such as an ARM architecture processor, is selected. This chip integrates a CPU, GPU, and dedicated NPU, enabling it to efficiently perform computationally intensive tasks such as image preprocessing, feature extraction, and classification comparison.

[0041] The processing module is equipped with a lightweight embedded operating system at the software level and runs the core recognition algorithm of this invention.

[0042] The display module uses a 3.5 to 5-inch capacitive touchscreen with a resolution of at least 720p. Its surface is coated with an anti-fingerprint and anti-glare coating to meet visibility requirements in various outdoor scenarios. This screen is used not only to display images and results but also serves as the primary human-computer interaction interface.

[0043] The power module uses a rechargeable lithium polymer battery with a typical capacity of 4000mAh, and uses a high-efficiency power management integrated circuit (PMIC) for charge and discharge management and voltage distribution to ensure stable system operation and provide sufficient offline working time.

[0044] The present invention also provides an insect identification method using an insect identification camera, comprising the following steps: Step 1: The image acquisition module acquires images of insects or wounds; Step 2: The processing module extracts the shape features or wound features from the image and compares the extracted features with the pre-stored database; Step 3: The display module outputs the insect type with the highest matching degree and processing suggestions.

[0045] Users select a recognition mode (shape or wound) via the touchscreen and complete the image capture. The processing module first preprocesses the raw image. This includes: Noise reduction: Use nonlocal means (NLM) or bilateral filtering algorithms to remove noise while preserving edge details.

[0046] Enhancement: The Limiting Contrast Adaptive Histogram Equalization (CLAHE) technique is used to improve local contrast in the image, making the texture of insects or the boundaries of wounds clearer.

[0047] Segmentation: For wound images, algorithms based on region growing or graph cut are used to separate the swollen area from the normal skin.

[0048] In shape recognition mode, the algorithm extracts features from preprocessed insect images. First, image segmentation is performed to separate the insect subject from the background, followed by the calculation of feature vectors. For example, the insect's outline is extracted and its Hu invariant moments are calculated to describe its shape; the image is converted to the HSV color space and a color histogram is calculated; keypoints and descriptors are extracted using SIFT or ORB algorithms to characterize local texture.

[0049] In wound recognition mode, the algorithm performs calculations on the segmented swollen area. The extracted features include: the equivalent diameter of the area, morphological features (such as convexity and compactness), color features (such as the mean R value in the RGB channels, the distribution of H and S in the HSV space), and texture features extracted through Gabor filter banks to quantify the uniformity of the swelling, the presence of blisters, etc.

[0050] This invention employs a dual-path parallel processing approach to extract feature signatures from the original image that can be used for comparison.

[0051] For shape recognition path: Traditional image features are used for rapid initial screening and to provide interpretability.

[0052] First, identify the shape characteristics of the insect. Use Hu invariant moments to describe the overall outline and shape of the insect. This feature is insensitive to scaling, rotation, and translation, making it ideal for identifying similar insects in different postures.

[0053] Then, the color characteristics of the insects are identified. Calculating the color histogram in the HSV color space, especially the H (hue) and S (saturation) components, can effectively characterize the color distribution and patterns on the insect's body (such as the yellow and black stripes of a bee).

[0054] Next, the texture features of the insects were identified. Local Binary Pattern (LBP) or Gabor filters were used to extract texture information of the insect wings and body surface, such as the grid structure of wing veins and the velvety texture of the body surface.

[0055] This invention also uses a pre-trained and pruned lightweight convolutional neural network (CNN), such as MobileNet, SqueezeNet, or ShuffleNet. These networks are designed for embedded devices, have few parameters, and are computationally efficient. The CNN does not directly classify the insect, but rather acts as a powerful "feature extractor." Insect images are input into the network, and a high-dimensional feature vector (e.g., a 512-dimensional vector) is extracted before the final fully connected layer. This vector integrates deep abstract information such as the insect's morphology, color, and texture, providing a representational capability far exceeding that of traditional features.

[0056] For wound recognition paths, traditional image processing algorithms are more relied upon to quantify morphological indicators of medical interest.

[0057] First, the wound area in the image is segmented: a region growing algorithm or a threshold-based segmentation algorithm is used to separate the swollen area from the normal skin in the image.

[0058] The segmented regions are then calculated, including the area, perimeter, equivalent diameter, and compactness (perimeter² / area).

[0059] It is also necessary to identify the shape characteristics of the wound. Hu invariant moments are also used to describe the overall shape of the swelling (whether it is circular, strip-shaped, or irregular).

[0060] The color and texture features of the wound are identified. The mean R (red) value of the swollen area in the RGB channel and the mean S (saturation) value in the HSV space are calculated to quantify the degree of redness and swelling.

[0061] It is also necessary to analyze whether there are clusters of bright or dark pixels in the center of the area to determine whether there are punctures or mouth parts remaining.

[0062] The uniformity of redness is characterized by using the statistical distribution of pixel values ​​(such as standard deviation).

[0063] The extracted features need to be matched against the database. The "Insect Morphological Feature Library" and "Insect Wound Correspondence Library" in the storage module do not actually store the original images, but rather a set of feature vector templates pre-extracted using the methods described above. Each template corresponds to an insect species label.

[0064] For traditional features, the K-Nearest Neighbor (K-NN) algorithm or Support Vector Machine (SVM) can be used. By calculating the Euclidean distance or cosine similarity between the feature vector of the image to be identified and the feature vectors of each template in the database, the closest "neighbor" is found, thereby determining the insect type.

[0065] Using cosine similarity for depth features is one of the most effective methods. This is because the orientation of depth feature vectors in high-dimensional space is more discriminative than their absolute value. The system calculates the cosine similarity between the depth features of the image to be identified and the depth features of each template in the database, and selects the one with the highest similarity.

[0066] Preferably, traditional features and deep features can be concatenated into a more comprehensive "fusion feature vector," which is then input into a linear SVM or a simple fully connected neural network for final classification. This classifier is pre-trained with a large amount of data before leaving the factory.

[0067] The decision-making and fusion mechanism is the "brain" of the algorithm, responsible for producing the final result.

[0068] When there is only one input pattern, such as only images of insects or only images of wounds, the system will directly output the result with the highest confidence level under that path.

[0069] When images of insects and wounds are presented simultaneously, the algorithm initiates a fusion decision. During the fusion decision process, the two recognition paths run independently, each providing a candidate result list containing the species and confidence level. If the results with the highest confidence levels from both paths are consistent, the system significantly improves the overall confidence level of the final result.

[0070] If the results of the two paths are inconsistent, in one implementation, the system will prioritize the result with higher confidence.

[0071] In one implementation, a preset rule base is enabled to directly report "information conflict" to the user and display two results and their confidence levels, requesting the user to provide a clearer image.

[0072] The rules in the rule base include: if the wound features strongly point to 'embedded bites', even if the appearance is identified as 'common beetle' with low confidence, it should be prioritized as 'tick' and the user should be prompted to check carefully.

[0073] The extracted feature vectors are compared with the feature data in the corresponding database. This embodiment uses one or a combination of the following two methods: Nearest Neighbor Classifier: Calculate the Euclidean distance or cosine similarity between the feature vector of the image to be identified and the feature vector of each sample in the database, and take the N samples with the smallest distance or the highest similarity as candidate results.

[0074] Lightweight Neural Network Classifier: A pre-trained, streamlined convolutional neural network (such as MobileNetV2 or variants of SqueezeNet) runs on an embedded NPU. This network takes a pre-processed image or feature map as input and directly outputs a probability distribution vector for the insect species. The system selects the category with the highest probability value as the recognition result.

[0075] The processing module sends the final identification results, including the insect name, Latin name, and matching confidence score, to the display module for display. Simultaneously, it also sends corresponding processing suggestions retrieved from the knowledge base to the display module. The results are presented in a graphical interface on the display module, and users can scroll through the touchscreen to view detailed information.

[0076] Treatment suggestions may include, for example, "Do not forcibly pull out ticks. Instead, use tweezers to hold the tick's head firmly against the skin and remove it vertically and evenly."

[0077] When a user provides both images of an insect and a wound, the system will initiate two recognition processes in parallel. The processing module will then combine the recognition results from both modes. For example, if the shape recognition result is "bee" (75% confidence), and the wound recognition result is also "bee sting" (90% confidence), the system will give a final judgment with higher overall confidence (e.g., 95%). If the two results are inconsistent, the system will prompt the user that "information may conflict" and suggest that the user retake a clearer image or prioritize the result with higher confidence.

[0078] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention.

[0079] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

[0080] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed by the present invention.

Claims

1. An insect identification camera, characterized in that, include: The image acquisition module includes a macro lens with an optical focal length of 3cm to 5cm and a fill light with a color temperature that can be steplessly adjusted in the range of 3000K to 6000K. The storage module is used to store a database of insect morphological features and a database corresponding to insect wounds; The processing module is used to run the recognition algorithm, extract image features, and compare them with the database; The display module is used to display images and recognition results; The power supply module provides power to each module.

2. The insect identification camera according to claim 1, characterized in that, The supplementary light includes multi-color temperature LEDs, and the brightness of each LED is independently controlled by pulse width modulation technology to achieve color temperature adjustment; the processing module is configured to automatically call up a supplementary light scheme of about 4500K in wound recognition mode and a supplementary light scheme of about 5500K in shape recognition mode.

3. The insect identification camera according to claim 1, characterized in that, The storage module uses an embedded multimedia card. The insect morphological feature library and the insect wound correspondence library are stored in a structured data format. The records include standard images, text descriptions, and feature vectors. The feature vectors include at least one of the following: shape descriptors based on Hu moments, color histogram data, local binary pattern texture features, and deep features extracted by a pre-trained convolutional neural network.

4. The insect identification camera according to claim 1, characterized in that, The processing module uses an embedded system-on-a-chip with a neural network processing unit and is equipped with a lightweight embedded operating system.

5. An insect identification method using an insect identification camera according to any one of claims 1-4, characterized in that, Includes the following steps: Images of insects or wounds are acquired using the image acquisition module; The processing module extracts the shape features or wound features from the image; The extracted features are compared with a pre-stored database of insect morphological features or a database of insect wound correspondences. The display module outputs the matching insect type and treatment suggestions.

6. The method according to claim 5, characterized in that, Before feature extraction, the image is preprocessed. The preprocessing includes at least one of the following operations: noise reduction using nonlocal mean or bilateral filtering algorithms, enhancement using contrast-limited adaptive histogram equalization, and segmentation of wound images using region growing or graph cut algorithms.

7. The method according to claim 5, characterized in that, In shape recognition mode, the extracted features include: Shape features described by calculating Hu invariant moments; Color histogram characteristics in the HSV color space; Texture features extracted using local binary patterns or Gabor filters; Deep features extracted using a lightweight convolutional neural network.

8. The method according to claim 5, characterized in that, In wound recognition mode, the extracted features include: At least one morphological feature among the area, perimeter, equivalent diameter, and compactness of the segmented inflamed region is calculated. Shape features described by Hu invariant moments; The swollen area exhibits at least one color characteristic in the mean R value of the RGB channel and the mean S value in the HSV space; The pixel analysis features in the central region are used to determine whether there are puncture holes or mouthpart remnants.

9. The method according to claim 5, characterized in that, The step of comparing the extracted features with the database includes: Calculate the Euclidean distance or cosine similarity between the feature vector of the image to be identified and the feature vector of the template in the database, and determine the insect type based on the K-nearest neighbor algorithm or support vector machine; and / or The image to be identified or its feature map is input into a lightweight neural network classifier to directly obtain the probability distribution of insect species.

10. The method according to claim 5, characterized in that, When both insect and wound images are input simultaneously, the processing module initiates the shape recognition and wound recognition processes in parallel, and combines the recognition results and confidence levels of the two processes. If the results are consistent, the overall confidence level is increased; if the results are inconsistent, a decision is made based on a preset rule base. The decision includes prioritizing the result with higher confidence or prompting the user with a conflicting information.