Malaria pathogen detection system based on improved YOLO algorithm

Abstract

The invention discloses a malaria pathogen detection system based on an improved YOLO algorithm. The malaria pathogen detection system comprises a blood sampling module, a microfluidic processing module, a microscopic imaging module, an AI analysis module and a display screen, the blood sampling module is used for quickly collecting a blood sample from a fingertip; the microfluidic processing module is used for integrating a thin film heater and accurately controlling the temperature of a sample; the microscopic imaging module is used for integrating a Raspberry Pi HQ Camera module and an automatic focusing motor MEMS (Micro Electro Mechanical System) micromirror, and converting a blood sample into a digital image through a high-precision lens and a dynamic focusing function; the AI analysis module is used for internally arranging an improved lightweight YOLO algorithm and carrying out efficient image processing and real-time pathogen detection in the equipment; and the display screen is used for outputting a real-time detection result and confidence in a two-stage display mode by using an OLED (Organic Light Emitting Diode) screen. According to the invention, the accuracy of disease detection is obviously improved.

Description

Technical Field

[0001] The present invention belongs to the technical fields of medical detection equipment and artificial intelligence assistance technology, and particularly relates to a malaria pathogen detection system based on an improved YOLO algorithm. Background Art

[0002] Malaria is one of the deadliest vector-borne infectious diseases globally. In the field of malaria detection, traditional microscopic detection methods rely on manual staining, microscopy, and pathogen identification. Although this method has high specificity, its practical application has significant drawbacks: High subjective error rate: Research shows that in scenarios with low parasite density (<100 parasites / μL) or mixed infections (such as simultaneous infection with Plasmodium vivax and Plasmodium falciparum), the misdiagnosis rate of manual microscopy can reach 15%-30%; Resource dependence restricts popularity: Microscopes, staining reagents, and stable power supply are difficult to guarantee in remote areas. Operational complexity and lack of timeliness: In areas with scarce medical resources, the training period for a qualified microscopy technician is as long as 6-12 months, resulting in a severe shortage of human resources; thus, it is difficult to meet the rapid screening needs of areas with scarce medical resources. In recent years, automated detection methods based on deep learning have gradually become a research hotspot, providing a new technical path for malaria detection.

[0003] Currently, object detection methods based on deep learning mainly focus on traditional models, such as algorithms like Faster-RCNN and CNN. Although these methods show high detection accuracy in certain scenarios, they still have significant limitations when dealing with the detection tasks of small targets and dense targets.

[0004] In terms of hardware implementation, the design of existing devices does not provide effective support for malaria detection. Most devices do not provide a dedicated mechanism for specific bottlenecks in the detection process (such as the integration of blood collection, imaging, and analysis), resulting in a cumbersome detection process and low efficiency, making it difficult to meet the actual needs of resource-constrained areas.

[0005] In recent years, some studies have attempted to use the YOLO framework for malaria detection. Although it shows certain advantages in real-time performance, the traditional YOLO framework still faces the following key bottlenecks: 1. Contradiction between model complexity and deployment: The existing YOLO series algorithms have a large number of parameters and high computational overhead, making it difficult to achieve real-time inference on portable edge devices.

[0006] 2. Insufficient detection accuracy for small targets: Malaria parasites are tiny and densely distributed. Traditional YOLO models are difficult to capture long-range global dependencies during feature extraction, resulting in low detection accuracy for small targets. In addition, the contrast between malaria parasites and the red blood cell background is low, further increasing the detection difficulty.

[0007] 3. Insufficient coordination between hardware and algorithm: The optimization of existing YOLO models mainly focuses on the algorithm level, without fully considering the co - design of hardware and algorithm.

[0008] 4. Feature extraction and background interference problems: When the backbone network of traditional YOLO models extracts features, it has insufficient ability to distinguish between malaria parasites and the background, and is easily interfered by background noise. This not only increases the false detection rate but also reduces the robustness of the model.

[0009] To address the above problems, the present invention proposes a portable malaria pathogen detector based on microscopic imaging and improved YOLO algorithm. Through the collaborative optimization of hardware and algorithm, it solves the bottleneck problems in malaria detection by traditional methods. Summary of the Invention

[0010] To solve the above - mentioned technical problems, the present invention proposes a malaria pathogen detection system based on an improved YOLO algorithm, which significantly improves the accuracy of disease detection.

[0011] To achieve the above object, the present invention provides a malaria pathogen detection system based on an improved YOLO algorithm, comprising: A blood collection module, a microfluidic processing module, a microscopic imaging module, an AI analysis module, and a display screen; The blood collection module is used to quickly collect blood samples from fingertips; The microfluidic processing module is used to integrate a thin - film heater to accurately control the temperature of the sample; The microscopic imaging module is used to integrate a Raspberry Pi HQ Camera module and an autofocus motor MEMS mirror. Through a high - precision lens and dynamic focusing function, it converts the blood sample into a digital image; The AI analysis module is used to build in an improved lightweight YOLO algorithm, perform lightweight optimization on the YOLO11 algorithm, and perform efficient image processing and real - time pathogen detection in the device; The display screen is used to use an OLED screen to output real - time detection results and confidence levels in a two - level display manner.

[0012] Technical effects of the present invention: (1) Optimize the YOLO11 model through an efficient channel attention module to enhance the feature extraction ability of the model, enabling the model to effectively focus on the key feature regions of malaria parasites in multi - scale images, thus significantly improving the accuracy of disease detection; (2) Utilize the global feature modeling ability of the transformer architecture to enhance the ability to detect small target malaria cells. This architecture can effectively capture the global dependencies between distant and tiny targets, further improving the recognition accuracy of small malaria cells and ensuring the efficient and accurate localization of malaria cells in the image. (3) Through lightweight convolutional blocks with bottlenecks, by simplifying the calculation process and optimizing parameters, the model significantly reduces the computational complexity while maintaining high accuracy. The application of this module effectively improves the speed of real-time inference, enabling the device to operate stably in a low-computation-resource environment and being suitable for portable devices. (4) Select DIoU-Loss as the loss function, comprehensively considering the overlapping area, center point distance, and scale relationship between the detection box and the ground truth box to optimize the target localization accuracy, reduce the offset of the detection box, and improve the stability and accuracy of pathogen detection.

[0013] (5) The detector realizes full-process automation. The detector of the present invention realizes full-process automation from sampling, staining, microscopic imaging to disease detection. This automated design not only reduces the complexity of manual operations but also ensures the efficiency and stability of each link, providing a convenient and low-threshold early malaria screening solution, especially suitable for initial self-screening and prevention in malaria-endemic areas. (6) Provide an efficient and low-cost preliminary solution for resource-constrained areas. By adopting the RaspberryPi HQ Camera module and the autofocus motor MEMS mirror, combined with modules such as an OLED display, a micro vacuum pump, a blood collection tube, and an integrated thin-film heater, a compact and efficient detection system is constructed. This enables the device to have the advantages of miniaturization, low power consumption, and low cost while ensuring accuracy and functionality. It provides an economical, practical, convenient, and efficient preliminary screening tool for primary medical institutions and home self-examinations, significantly reducing medical costs and enhancing the accessibility of malaria prevention and control. Brief Description of the Drawings

[0014] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation of this application. In the drawings: Figure 1 It is a framework diagram of the improved YOLO11 model according to an embodiment of the present invention; Figure 2 It is a schematic diagram of a two-branch convolutional layer module with attention added according to an embodiment of the present invention; Figure 3 It is a schematic diagram of a convolutional layer module with attention added according to an embodiment of the present invention; Figure 4Schematic diagram of the two-branch convolutional layer module with a transformer in the embodiment of the present invention; Figure 5 Schematic diagram of the convolutional layer module with a transformer in the embodiment of the present invention; Figure 6 Schematic diagram of the structure of the malaria pathogen detection system based on the improved YOLO algorithm in the embodiment of the present invention; Figure 7 Internal structure schematic diagram of the structure in the embodiment of the present invention; Among them, 1 - blood collection module, 2 - microfluidic processing module, 3 - microscopic imaging module, 4 - AI analysis module, and 5 - display screen, 6 - blood collection needle, 7 - blood collection tube, 8 - anticoagulation treatment cavity, 9 - controllable valve, 10 - micro vacuum pump, 11 - integrated thin film heating device, 12 - centrifugal separation area, 13 - staining reaction pool, 14 - Raspberry Pi HQ Camera module, 15 - lighting system, 16 - recovery cavity. Detailed implementation manners

[0015] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0016] It should be noted that the steps shown in the flowchart of the drawings may be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than here.

[0017] As Figures 1 - 7 shown, in this embodiment, a malaria pathogen detection system based on the improved YOLO algorithm is provided, including: a blood collection module 1, a microfluidic processing module 2, a microscopic imaging module 3, an AI analysis module 4, and a display screen 5.

[0018] Blood collection module 1: It includes a needle, a micro vacuum pump, and a blood collection tube, and is designed to quickly collect blood samples from the fingertip. The suction of the micro vacuum pump is adjusted through a pressure sensor to ensure that the blood collection process is simple and efficient; Microfluidic processing module 2: It integrates a thin film heater to precisely control the temperature of the sample. At the same time, it includes a centrifugal separator and a staining reactor. The former is used to remove impurities in the blood, and the latter is used to store and use Giemsa staining solution to stain the blood sample. This microfluidic design improves the automation and efficiency of sample processing; Microscopic imaging module 3: It integrates a Raspberry Pi HQ Camera module and an autofocus motor MEMS mirror. Through a high-precision lens and a dynamic focusing function, the blood sample is converted into a digital image to ensure that the imaging process is fast and clear; Detection module 4: It incorporates an improved lightweight YOLO algorithm, which has been optimized for lightweight processing of the YOLO11 algorithm, enabling efficient image processing and real-time pathogen detection within the device. Display module 5: An OLED screen is used to output real-time detection results and confidence levels in a two-level display manner. The first level shows the detection results, and the second level shows the detection confidence level.

[0019] Furthermore, as Figure 1 shown, the improved lightweight YOLO algorithm includes: Combining the two-branch convolutional layer modules in the backbone network of the efficient channel attention module; Fusing the transformer architecture into the backbone network of YOLO11; Introducing convolutional blocks with bottlenecks into the neck network; Using DIoU-Loss as the loss function.

[0020] Furthermore, combining the two-branch convolutional layer modules in the backbone network of the efficient channel attention module includes: For the input malaria information , initially extract the original features through two layers of convolutional operations , where: ; Subsequently, the extracted features pass through an improved two-branch convolutional layer module with attention to obtain malaria features .

[0021] The two-branch convolutional layer module with attention is an improved variant of the two-branch convolutional layer module. The core improvements include: fixing the internal structure of the two-branch convolutional layer module as a convolutional layer module and simultaneously introducing the efficient channel attention module. The calculation process of the two-branch convolutional layer module with attention can be expressed as follows: ; where, is the feature output of the nth convolutional layer module; represents the input of the nth two-branch convolutional layer module with attention; represents the output after passing through the nth two-branch convolutional layer module with attention; represents the output of the first two-branch convolutional layer module with attention. Among them, the efficient channel attention uses a band matrix to learn channel attention: ; where CID represents one-dimensional convolution.

[0022] Furthermore, fusing the transformer architecture into the backbone network of YOLO11 includes: ; where represents the output of the nth convolutional layer module with an added transformer; represents the output of the two-branch convolutional layer module with an added transformer; is the transformer architecture module; is the output of the two-branch convolutional layer module.

[0023] Furthermore, a convolutional block with a bottleneck is introduced into the neck network, including: Using a convolutional block with a bottleneck to replace the two-branch convolutional layer module in the YOLO11 neck network, and replacing the original two-branch convolutional layer structure with a multi-dilation rate cascaded convolutional block with a bottleneck, so as to effectively expand the receptive field and perform hierarchical extraction of cross-scale morphological features.

[0024] Furthermore, using DIoU-Loss as the loss function, including: ; where and respectively represent the center points of the predicted box and the ground truth box, represents the Euclidean distance between the two center points, represents the diagonal distance of the minimum enclosing region of the predicted box and the ground truth box. Among them, can be defined as: .

[0025] The specific structure of the blood collection module 1 is as follows: As Figures 6 - 7 shown, the blood collection module 1 includes a blood collection needle 6, a blood collection tube 7, an anticoagulation treatment cavity 8, and a micro vacuum pump 9. According to medical requirements, the blood collection needle 6 is a replaceable structure and needs to be replaced after each use. And the blood collection needle 6 and the detection host are detachably connected, which is convenient for timely replacement after each use. The anticoagulation treatment cavity 8 adopts a left-right double-chamber design. The left chamber is used as the blood collection and temporary storage area, and the right chamber is used as the anticoagulant mixing and treatment area. The blood collection needle 6 is connected to the left chamber inlet through the blood collection tube 7. When the blood in the left chamber reaches the preset volume, the controllable valve 9 set in the middle is automatically opened, and the blood flows into the right chamber. The right chamber is internally provided with anticoagulant, and the micro vacuum pump 10 provides power to transport the blood and make it fully mixed with the anticoagulant. After the blood is processed, it enters the microfluidic processing module 2 through the right chamber outlet.

[0026] The specific structure of the microfluidic processing module 2 is as follows: As Figures 6 - 7As shown in the figure, the microfluidic processing module 2 includes an integrated thin-film heating device 11, a centrifugal separation zone 12, and a staining reaction pool 13. The integrated thin-film heating device 11 preheats each functional zone to 37°C, and the separation of blood cells and plasma is completed through the centrifugal separation zone 12. After the pre-stored dry Giemsa staining solution is completely dissolved, the target components enter the staining reaction pool 13, and the stained samples will be automatically transmitted to the microscopic imaging module 3; the remaining components directly enter the recovery chamber 16.

[0027] The specific structure of the microscopic imaging module 3 is as follows: As Figures 6 - 7 shown in the figure, the microscopic imaging module 3 includes a RaspberryPi HQ Camera module 14 and an illumination system 15, which can realize the acquisition of high-resolution images and automatic focusing operations. The malaria cell images obtained by the microscopic imaging module 3 are transmitted into the AI analysis module 4. This module performs image recognition and malaria pathogen detection based on the improved YOLO11 model, and synchronizes the analysis results to the display screen 5. The display content is divided into two levels: the first level is the detection result (whether malaria is infected), and the second level is the recognition confidence score. The detected blood samples enter the detachable recovery chamber 16. This design facilitates cleaning and replacement, and helps to ensure the cleanliness of the whole machine and the convenience of repeated use.

[0028] The embodiment of the present invention provides a design method for a malaria pathogen detector based on an improved YOLO algorithm, aiming to fill the application gap of deep learning models in the field of malaria detection and solve practical problems such as low detection accuracy, insufficient medical resources, and difficulty for individuals to achieve early self-screening in malaria-high and resource-scarce areas. Among them, for the deep learning model, we choose to use the latest YOLO11 model. For general detection problems, the present invention introduces an efficient channel attention module to enhance the model's perception ability of the key features of malaria cells, so as to effectively distinguish pathogens from complex backgrounds in the image, and integrates the transformer architecture to construct global feature expression ability, enhancing the detection and recognition effects of the model on malaria cells with long-distance distribution and small size in the image. At the same time, DIoU-Loss is selected as the target localization loss function to optimize the matching accuracy between the prediction box and the ground truth box, reduce the target box offset, and integrate lightweight convolutional blocks with bottlenecks in the neck network. On the premise of ensuring the detection performance, the computational complexity is significantly reduced, and the inference efficiency and deployment feasibility of the model in edge computing devices are improved. This method has good versatility and is applicable to single-target and multi-target detection tasks, covering multiple fields such as artificial intelligence and medical detection, providing an efficient and innovative technical path for the practical application of medical image recognition.

[0029] The above are only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A malaria pathogen detection system based on an improved YOLO algorithm, characterized in that, Including: A blood collection module (1), a microfluidic processing module (2), a microscopic imaging module (3), an AI analysis module (4), and a display screen (5); The blood collection module (1) is used to quickly collect a blood sample from a fingertip; The microfluidic processing module (2) is used to integrate a thin film heater to precisely control the temperature of the sample; The microscopic imaging module (3) is used to integrate a Raspberry Pi HQ Camera module and an autofocus motor MEMS mirror, and through a high-precision lens and a dynamic focusing function, convert the blood sample into a digital image; The AI analysis module (4) is used to build in an improved lightweight YOLO algorithm, perform lightweight optimization on the YOLO11 algorithm, and perform efficient image processing and real-time pathogen detection in the device; The display screen (5) is used to use an OLED screen to output real-time detection results and confidence levels in a two-level display manner.

2. The malaria pathogen detection system based on the improved YOLO algorithm according to claim 1, characterized in that, The improved lightweight YOLO algorithm includes: Combining the two-branch convolutional layer modules in the backbone network of the efficient channel attention module; Fusing the transformer architecture into the backbone network of YOLO11; Introducing a convolutional block with a bottleneck into the neck network; Using DIoU-Loss as the loss function.

3. The malaria pathogen detection system based on the improved YOLO algorithm according to claim 2, wherein, Combining the two-branch convolutional layer modules in the backbone network of the efficient channel attention module, including: For the input malaria information , the original features are first extracted through two layers of convolutional operations , where: ； The extracted features pass through an improved attention-added two-branch convolutional layer module to obtain malaria features .

4. The malaria pathogen detection system based on the improved YOLO algorithm according to claim 2, characterized in that, The convolutional layer module with added attention is an improved variant of the two-branch convolutional layer module. The core improvement points include: fixing the internal structure of the two-branch convolutional layer module as a convolutional layer module, and at the same time introducing an efficient channel attention module. The calculation process of the convolutional layer module with added attention can be expressed as follows: ； Among them, is the feature output of the nth convolutional layer module; represents the input of the nth convolutional layer module with added attention; represents the output after passing through the nth convolutional layer module with added attention; represents the output of the first two-branch convolutional layer module with added attention.

5. The malaria pathogen detection system based on the improved YOLO algorithm according to claim 2, characterized in that, Fusing the transformer architecture into the backbone network of YOLO11, including: ； Among them, represents the output of the nth convolutional layer module of the addition transformer; represents the output of the two-branch convolutional layer module of the addition transformer; is the transformer architecture module; is the output of the two-branch convolutional layer module.

6. The malaria pathogen detection system based on the improved YOLO algorithm according to claim 2, characterized in that, Introducing a convolutional block with a bottleneck into the neck network, including: Using a convolutional block with a bottleneck to replace the two-branch convolutional layer module in the YOLO11 neck network, and replacing the original two-branch convolutional layer structure with a multi-dilation rate cascaded convolutional block with a bottleneck, so as to effectively expand the receptive field and perform hierarchical extraction of cross-scale morphological features.

7. The malaria pathogen detection system based on the improved YOLO algorithm according to claim 2, characterized in that, Using DIoU-Loss as the loss function, including: ； wherein and respectively represent the center points of the predicted box and the ground truth box, represents the Euclidean distance between the two center points, represents the diagonal distance of the minimum bounding region of the predicted box and the ground truth box.

Images