Insect specimen information intelligent acquisition and identification system based on deep learning

Abstract

The invention discloses an insect specimen information intelligent acquisition and identification system based on deep learning. The insect specimen information intelligent acquisition and identification system comprises a high-throughput automatic imaging system and an AI auxiliary insect identification expert system, the high-throughput automatic imaging system comprises a target detection system, an automatic control system and a photographing system. The AI-assisted insect identification expert system comprises an image database, an AI-driven insect specimen and field photo mixed identification system and an AI-assisted insect identification platform. According to the invention, an AI deep learning model is creatively combined with a traditional taxonomy expert system, an innovative insect identification and classification system is built, and a multifunctional community platform is expected to be built.

Description

Technical Field

[0001] This invention belongs to the field of taxonomy technology, specifically relating to an intelligent collection and identification system for insect specimen information based on deep learning. Background Technology

[0002] As people's awareness of environmental protection continues to rise, the conservation of animal biodiversity, as an important starting point and indicator of environmental protection, is receiving increasing attention from experts, scholars, and the public. In the animal kingdom, insects, with their rich diversity of over one million species, play crucial roles in areas such as plant pollination and decomposition, and their vibrant and unique colors and forms have garnered significant attention from academics and enthusiasts.

[0003] In entomological research and conservation, the digitization and identification of insect specimens are crucial. The ability to quickly digitize and identify insect specimens is of great importance, whether for the rapid collection of insect specimens in herbaria or for biodiversity statistics during biological surveys.

[0004] However, existing insect specimen digitization equipment has many problems. For example, the utility model patents applied for by patent numbers CN202420119465.8 and CN202223284092.X suffer from low imaging efficiency, relying on manual photography and unable to achieve batch processing. This makes it difficult to acquire high-quality images of a large number of specimens in a short time, significantly limiting research progress when dealing with massive numbers of specimens. Furthermore, due to the limitations imposed by insect color and size, conventional photographic equipment often requires removing the insects individually for photography, which undoubtedly causes some damage to the specimens. Automated technology, if applied to photographic devices, would allow for batch photography simply by removing the entire specimen box. Furthermore, target detection models based on deep learning algorithms, such as the commonly used YOLO, can quickly detect the target location of insect specimens in the specimen box, pinpoint the specimen's location, output relevant parameters, automatically control the camera to the appropriate position, and, after adjustment, automatically and in batches photograph the specimens. In scenarios such as herbariums where large-scale specimen photography and database creation are required, the entire process can be automated, saving human resources, time resources, and economic costs.

[0005] Furthermore, mainstream AI identification tools have significant shortcomings in insect identification. Their accuracy falls short of the stringent requirements of taxonomy, especially when dealing with rare species with limited data or closely related species with similar morphology, leading to frequent misclassifications. Moreover, most of these tools can only process specimen images or field ecological images individually, failing to support combined identification of both, as illustrated in patent application CN202420408106.4. This makes it difficult for researchers to utilize AI tools for comprehensive analysis when combining specimen studies with field observations, limiting the depth and breadth of insect research. More rigorous expert identification methods require highly experienced specialists, often with limited resources available. Therefore, there is an urgent need to develop more efficient, accurate, and compatible insect specimen identification technologies to advance entomology and promote insect diversity conservation. Classical taxonomy methods can provide us with high accuracy in identifying rare and closely related species, while species classification models based on deep learning algorithms (such as ResNet residual neural networks) can provide rapid identification for most common species. Combining these methods not only inherits the rigor of traditional taxonomy but also leverages the powerful computing capabilities of AI technology to significantly improve the accuracy of insect identification, effectively recognizing morphologically similar insect species. Summary of the Invention

[0006] To overcome the shortcomings of existing technologies, this invention provides a deep learning-based intelligent acquisition and identification system for insect specimens, including a high-throughput automatic imaging system and an AI-assisted insect identification expert system. The high-throughput automatic imaging system includes a target detection system, an automatic control system, and an imaging system. The AI-assisted insect identification expert system includes an image database, an AI-driven hybrid recognition system for insect specimens and field photographs, and an AI-assisted insect identification platform. This invention innovatively combines AI's deep learning model with a traditional taxonomic expert system to build an innovative insect identification and classification system, and aims to construct a multifunctional community platform.

[0007] The technical solution adopted by this invention to solve its technical problem is as follows: A deep learning-based intelligent acquisition and identification system for insect specimen information, comprising a high-throughput automatic imaging system and an AI-assisted insect identification expert system; The high-throughput automatic imaging system includes a target detection system, an automatic control system, and an imaging system; The target detection system uses a YOLO-based deep learning computer vision algorithm. It uses a host computer to input images into a processor for target detection, obtains the coordinates of the area to be photographed and the current camera coordinates, and transmits the route to the automatic control system. The automatic control system uses coordinates and routes to control the stepper motor to drive the lead screw and transmission belt, thereby enabling the gimbal with camera to move sequentially to the position directly above all targets and take pictures. When the camera reaches the appropriate position, the photography system confirms that all parameters for taking the picture are correct and starts the camera shutter. The AI-assisted insect identification expert system includes an image database, an AI-driven system for recognizing a mixture of insect specimens and field photographs, and an AI-assisted insect identification platform. The image database consists of images collected using a high-throughput automatic imaging system, providing support for an AI-driven hybrid recognition system that combines insect specimens with field photographs. The AI-driven hybrid recognition system for insect specimens and field photos is based on the construction and training of a ResNet residual neural network. Adjustments are made to the pre-trained model, and the fully connected layers are frozen. During the model training process, parameters are adjusted to optimize the model, regularization is applied to prevent overfitting, and an image database is used for training. The AI-assisted insect identification platform includes a front-end interface, a back-end server, a database, and an AI model module; it supports user input of morphological features and ecological information for assisted identification; and it deeply integrates the AI ​​recognition system with a taxonomic expert system to achieve joint identification functionality.

[0008] Preferably, the images in the image database are images of insect specimens, covering different insect orders, families, and genera.

[0009] Preferably, the images in the image database are classified, labeled, and managed.

[0010] Preferably, the images in the image database are preprocessed, including image cropping, normalization, and enhancement operations.

[0011] The beneficial effects of this invention are as follows: (1) The present invention proposes a protective batch direct imaging method, which can avoid the shortened specimen life caused by conventional photography.

[0012] (2) Current insect imaging technology mainly relies on human labor. This invention has the advantages of high efficiency, high quality and batch imaging.

[0013] (3) The present invention has high scalability and can be modified and expanded according to specific needs, making it highly applicable.

[0014] (4) This invention innovatively combines AI's deep learning model with traditional taxonomy expert systems to build an innovative insect identification and classification system, and aims to construct a multifunctional community platform. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a high-throughput automatic imaging system. Figure 2 Design drawings for part of the automatic control system; Figure 3 Schematic diagram of the overall technical route of this invention; Figure 4 This is the key content of the research in this invention; Figure 5 Three-view diagram of a high-throughput automatic imaging system; Figure 6 This is a prototype of a high-throughput automatic imaging system. Detailed Implementation

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] The present invention provides an intelligent acquisition and identification system for insect specimen information based on deep learning, which solves the problems of low shooting efficiency, potential damage to specimen resources, low accuracy in identifying specific species, and insufficient identification expert resources of existing methods.

[0018] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: (1) Develop a high-throughput automatic imaging system (such as...) Figure 1 This system can efficiently and accurately acquire high-quality images of insect specimens, significantly improving imaging efficiency.

[0019] (2) By integrating classical taxonomic methods, an AI-assisted insect identification expert system was constructed based on deep learning algorithms.

[0020] The technical solution is as follows: (1) The overall technical route is as follows Figure 3 ; This invention is divided into two main sections, which are interconnected. The first section is the construction of a high-throughput automatic imaging system, which mainly consists of three modules: a target detection system, an automatic control system, and an imaging system (the specific implementation idea is as follows). Figure 2The target detection system primarily relies on YOLO-based deep learning computer vision algorithms. The host computer of the device transmits images to the processor for target detection, obtaining the coordinates of the targets to be photographed and the current camera coordinates, and then transmits the route to the automatic control system. The automatic control system uses the coordinates and route to control a stepper motor that drives a lead screw and transmission belt, thereby enabling the pan-tilt head with the camera to sequentially move to the appropriate position directly above all targets and take pictures. The imaging system mainly confirms that all parameters are correct when the camera reaches the appropriate position and then activates the camera shutter. After the high-throughput automatic imaging system is completed, it can be used to collaborate with relevant units and departments to take a large number of insect specimen images, as well as obtain images through other means, to provide necessary support for the second part.

[0021] The second part is the construction of a classification and identification platform, which mainly consists of three parts. The first part involves establishing a relevant database using collected images to support the training of the AI ​​model for the classification system. The second part is the construction and training of a ResNet-based residual neural network. This invention adjusts the pre-trained model by freezing the fully connected layers and replacing the remaining layers. During model training, parameter adjustments and optimizations are performed, regularization is applied to prevent overfitting, and images are used for training, completing the AI ​​model for classification and identification. The third part is the expert system component, which integrates with the classification system to achieve an "AI-assisted expert system" classification and identification mode. Plans are also underway to incorporate GIS and other systems to expand functionality. After integrating these three parts, the classification and identification platform of the second part of this invention is successfully built. The two parts can then work together to form a one-stop solution for the intelligent collection and classification of insect specimens in batches.

[0022] 2) Key contents of this invention, such as Figure 4 : (1) Construction of insect specimen photography system and image database; The initial phase involved building a high-throughput automated imaging system and establishing an insect specimen image database to store and manage the acquired image data. The high-throughput automated imaging system, including an optical imaging module, an automatic control module, and a data acquisition module, was designed and built. First, design drawings of the imaging system were created, and a model was built. After the model was successfully built, the automated imaging system was constructed. Insect specimen images were acquired simultaneously, ensuring that the images covered as many different insect orders, families, and genera as possible. An image database was developed to support subsequent functionalities. The acquired image data was imported into the database and then classified, labeled, and managed.

[0023] (2) Development of an AI-driven hybrid recognition system for insect specimens and field photos; Building upon previous work, we will develop an AI recognition system compatible with both specimen and field ecology images to improve insect identification accuracy. We will collect and organize insect specimen images and field ecology photos from the Qinling Mountains to construct a large-scale training dataset. Data preprocessing will be performed, including image cropping, normalization, and enhancement. Suitable deep learning frameworks (such as CNN and ResNet) for insect image recognition will be selected. We will develop a system that supports joint recognition of specimen and field ecology images. We plan to optimize model performance and improve recognition accuracy through techniques such as transfer learning and data augmentation. The trained AI model will be integrated into the recognition system or the research group's website, and a user-friendly interface will be developed. Functions such as image uploading, displaying recognition results, and recommending similar species will be implemented.

[0024] (3) Construction of an AI-assisted insect identification platform that integrates classic insect classification and identification methods; This project aims to build an insect identification platform that combines AI assistance with an expert system, integrating classical taxonomic methods to provide more comprehensive identification support. The overall platform architecture was designed, including a front-end interface, back-end server, database, and AI model module. A classical taxonomic identification module was developed, supporting user input of morphological features and ecological information for assisted identification. The AI ​​recognition system was deeply integrated with the classical taxonomic expert system to achieve joint identification functionality. Systems such as GIS were also incorporated to enrich its functionality. Functional testing and optimization were conducted to ensure stable system operation. Future updates will optimize platform functionality based on user feedback to improve the user experience.

[0025] like Figure 5 As shown in the three-view drawing, the device is constructed from several profiles with lengths of 400mm, 360mm, and 370mm (specific dimensions can be adjusted as needed). Three stepper motors control the X, Y, and Z axes respectively. The X-axis stepper motor drives a belt connected to the stage, thus moving the stage along its X-axis. The Y-axis stepper motor moves the gimbal equipped with pulleys via a belt drive. The Z-axis movement is achieved by connecting a stepper motor to a lead screw, which in turn moves the stage along its Z-axis. A detailed theoretical prototype is shown below. Figure 6 As shown.

[0026] This invention utilizes a photographic system to collect specimens and collaborates with relevant organizations to establish a related specimen database. It aims to provide relevant websites or software for entomological researchers and enthusiasts, thereby contributing to the widespread promotion of insect research and popular science.

[0027] Computer technology can be used to collect information such as the location of insects while collecting photos of them. The collected information can be calculated and simulated to obtain information such as the distribution of species and the invasion of alien species, which can help with species surveys and environmental protection.

Claims

1. A deep learning-based intelligent acquisition and identification system for insect specimen information, characterized in that, This includes high-throughput automated imaging systems and AI-assisted insect identification expert systems; The high-throughput automatic imaging system includes a target detection system, an automatic control system, and an imaging system; The target detection system uses a YOLO-based deep learning computer vision algorithm. It uses a host computer to input images into a processor for target detection, obtains the coordinates of the area to be photographed and the current camera coordinates, and transmits the route to the automatic control system. The automatic control system uses coordinates and routes to control the stepper motor to drive the lead screw and transmission belt, thereby enabling the gimbal with camera to move sequentially to the position directly above all targets and take pictures. When the camera reaches the appropriate position, the photography system confirms that all parameters for taking the picture are correct and starts the camera shutter. The AI-assisted insect identification expert system includes an image database, an AI-driven system for recognizing a mixture of insect specimens and field photographs, and an AI-assisted insect identification platform. The image database consists of images collected using a high-throughput automatic imaging system, providing support for an AI-driven hybrid recognition system that combines insect specimens with field photographs. The AI-driven hybrid recognition system for insect specimens and field photos is based on the construction and training of a ResNet residual neural network. Adjustments are made to the pre-trained model, and the fully connected layers are frozen. During the model training process, parameters are adjusted to optimize the model, regularization is applied to prevent overfitting, and an image database is used for training. The AI-assisted insect identification platform includes a front-end interface, a back-end server, a database, and an AI model module; it supports user input of morphological features and ecological information for assisted identification; and it deeply integrates the AI ​​recognition system with a taxonomic expert system to achieve joint identification functionality.

2. The intelligent acquisition and identification system for insect specimen information based on deep learning according to claim 1, characterized in that, The image database contains images of insect specimens, covering different insect orders, families, and genera.

3. The intelligent acquisition and identification system for insect specimen information based on deep learning according to claim 1, characterized in that, The images in the image database are classified, labeled, and managed.

4. The intelligent acquisition and identification system for insect specimen information based on deep learning according to claim 1, characterized in that, The images in the image database are preprocessed, including image cropping, normalization, and enhancement operations.

Images