YOLOv12-Based Parasitic Egg Detection for Caprine Diagnostics

YOLOv12-Based Parasitic Egg Detection for Caprine Diagnostics

B. Lakshmi Narayana Reddy

Department of ECE, Sri Venkateswara College of Engineering, Tirupati, A.P., India

Hasti Nihaarika

Department of ECE, Sri Venkateswara College of Engineering, Tirupati, A.P., India

C. Raju

Department of ECE, Sri venkateswara college of Engineering, Tirupati, A.P., India

Cherukuru Jaswanth

Department of ECE, Sri Venkateswara College of Engineering, Tirupati, A.P., India

Kandavel Pavan Kumar

Department of ECE, Sri venkateswara college of Engineering, Tirupati, A.P., India

Byreddy Yudhister Reddy

Department of ECE, Sri venkateswara college of Engineering, Tirupati, A.P., India

A Chenna Kesava Reddy

Department of ECE, Sri Venkateswara College of Engineering, Tirupati, A.P., India

Abstract – Caprine inhabitants also face parasitic infection which greatly affects the health of the livestock, resulting in decreased productivity and more veterinary treatment. Parasitic eggs need to be detected and correctly classified in good time to help control and prevent the disease. The present paper presents a deep learning-based model based on the YOLOv12 (You Only Look Once) model to automatize the process of parasitic egg detection and classification in goats. The research uses a wide set of data of 11 types of parasitic eggs, such as Ascaris lumbricoides, Capillaria philippinensis, Enterobius vermicularis, Fasciolopsis buski, Hookworm eggs, Hymenolepis diminuta, Hymenolepis nana, Opisthorchis viverrine, Paragonimus spp., Taenia spp. egg, and Trichuris trichiura, and has 1000 images of each. YOLOv12 is known to be one of the state of the art object detectors that are trained to identify these parasitic types of eggs effectively. Accuracy, precision, recall, F1 score are the common measurements to assess the performance of the model which indicates that it can be utilized in veterinary diagnostics. The paper will present a new idea of how to automatize the process of detection of parasites in goats to shorten the time spent on a diagnosis and enhance the management of intervention. This system may help the veterinarians in treating parasitic diseases because it will enable them to identify them fast, thereby improving the health of caprines and reducing the economic effects of parasitic infections in livestock. The study is a necessary step in the development of automated diagnostic tools of parasitic infections and the opportunities of the further use in the animal health management and accuracy in veterinary care.

Keywords: Caprine health, YOLOv12, machine learning, parasitic egg detection, veterinary diagnostics, deep learning, livestock management, image classification.

1. INTRODUCTION

   Goat parasitism is a threatening health, production, and quality of livestock. These parasites have the ability to slow down growth rates, diminish reproductive performance and raise

   veterinary expenses since goats are very prone to many parasitism infections. The conventional diagnostic procedures such as the manual microscopic analysis on fecal samples are both time consuming and labor intensive and also subject to human error; thus, early diagnosis and intervention becomes a challenge.

   Recent breakthroughs in computer vision and deep learning provide opportunities to implement automation of the process of detecting and classifying parasitic eggs in livestock. The newest version of the YOLO (You Only Look Once) algorithms, the YOLOv12, is excellent in detecting and categorizing objects in real-time in images. YOLOv12 is a suitable tool that can be used in veterinary diagnostics due to its enhanced accuracy and speed.

   The proposed paper is a deep learning model based on YOLOv12, which will be used to detect and classify 11 different types of parasitic eggs in goat fecal samples. With the help of a special database of parasitic egg pictures, the model will be able to offer correct, quick, and dependable detection, which is going to decrease diagnostic time and raise the efficiency of interventions significantly. The adoption of such a system may revolutionize the field of veterinary diagnostics, as it will render them faster and more accurate, which will eventually enhance the well-being of animals and ensure a sustainable production of livestock.

   1. Background

      Caprine parasitic infections are health and economically threatening to the populations of ruminants. The conventional methods of diagnosis, such as manual microscopic analysis of fecal samples, are tedious and are subject to human error and hence delayed treatments. As the level of requirement in this regard increases, advanced deep learning algorithms such as YOLOv12 are developed. This is an improved real time object detection model with faster, automatic and highly accurate classification of parasitic eggs to facilitate faster diagnostics and early intervention in veterinary care which will eventually lead to better health of animals and minimal outbreaks.

   2. Problem Statement

      The health and farm production of goats is greatly impacted by parasitic infection. Early diagnosis is paramount, however, the conventional approach, such as fecal egg count, is time-consuming, needs human skills, and is largely subjective, with diagnoses made late. This leads to infestation in the long term, expensive veterinary care and inefficiency especially in large farms where prompt intervention is crucial. The existing diagnostic devices are not automated, which restricts the ability to engage in proactive disease control, particularly in the areas where the quality of the veterinary services is low.

      Due to the rising number of parasitic diseases, there is a great demand to have an automated system with high capacity capable of detecting parasitic eggs rapidly and with high accuracy. Recent research has indicated that deep learning may have the solutions to this disease detection using images, but limited research has been carried out on the application of deep learning to parasitic egg detection among goats.

      The gap that is addressed by this paper is through the use of YOLOv12, which is an effective deep learning object detector. We have attempted to make a potent system that can detect parasitic eggs of different varieties with high accuracy and speed by training YOLOv12 on a dataset of parasitic eggs in goats. This solution defeats the shortcomings of the traditional approaches, providing fast, scalable and precise diagnostics. Installation of this system will result in increased health of the animals, less expenditure of the veterinary services and increased productivity of the farm as a whole.

   3. Objectives and Contributions

      The primary objective of the provided work is to develop a deep learning system using the assistance of the YOLOv12 model that will be capable of automatic detection and classification of parasitic eggs in goats. It is aimed at achieving high accuracy in the definition of various types of parasites with the entire set of 11 groups of parasitic eggs conditioned on a large-scale data set, which would allow diagnosing the disease and proceed to take appropriate precautions in time. This mechanism will offer an effective alternative to the traditional methods of diagnosing the parasitic diseases amongst livestock, and this will save a lot of time and labor used in this procedure.

      The research addition value lies in the fact that the state- of-the-art object detection techniques were implemented in veterinary diagnostics. The research makes the way to developing the system where it is possible to achieve the realization of automated diagnostic system in veterinary practice, as it demonstrates the possibility of the use of the

      YOLOv12 in the classification of parasitic eggs. The proposed model can be also scaled to applications on users in the future to track the livestock health to a greater scale to improve the disease management plans and foster sustainable farming practices within the global market environment.

2. REVIEW OF LITERATURE

   Deep learning-based detection of parasitic eggs has recently become one of the key research fields because of the weaknesses of manual microscopy diagnostics and the necessity of prompt and correct diagnosis. The original dataset on the ICIP 2022 Challenge on Parasitic Egg Detection and Classification in Microscopic Images was used as a benchmark and an evaluation criterion of parasite egg detection, and various object detection algorithms were designed to use convolutional neural networks (CNNs) and genetic architectures to classify parasite eggs in microscopic smear images [1].

   The transformer-based architectures in detecting parasitic eggs as part of the ICIP challenge. It implies that, even in the context of a single CNN paradigm, additional feature representation richness, and more local features, can enhance detection pipelines [2]. Another strength of ensemble and multimodel methods, including multiscale weighted fusion and ensemble learning, was also emphasized as useful in the ICIP challenge, as they resulted in higher accuracy in classifying types of eggs and showed the benefit of applying numerous detectors to improve robustness to image variability [3].

   Based on the challenges in the ICIP benchmark, lightweight deep learning models that are microscopy-friendly have been suggested to overcome the constraints of computational resources in resource-constrained settings. An example to this includes the YAC-Net architecture that fuses adaptive features, and modified backbone modules to enable the system to produce high-precision and recall at a cost of lowering network complexity in detecting parasitic eggs [4]. The models state that accuracy and efficiency have to be balanced particularly in field diagnostics, when the laboratory is low-resource.

   Subsequent studies on ensemble models and optimization techniques have resulted in successful systems. Wang et al. created an ensemble system which integrates the YOLOv5 with Cascade R-CNN to enhance the detection sensitivity with the use of data augmentation and transfer learning to reduce the false negatives and increase the generalizability of the system to various microscopic conditions [5]. Wan et al. likewise introduced a C2BNet model, which has a composite backbone, which facilitates multiscale feature mining and thereby resulting in the accurate detection of parasitic eggs despite various factors such as image blur or change of focus [6].

   Certain models such as YOLOv3 and other one-stage detectors have been fundamental in real-time object classification tasks in objects detection research, due to their provision of both speed and accuracy, which is the reason why they have been successfully used in parasitology and biomedical imaging [7]. The state-of-the-art object detectors have progressively been refined over the years and have advanced in automatic classification of objects in various

   applications like in medical imaging and agricultural automation [8].

   Also, recent advances in data preparation and data augmentation techniques have helped to improve the detection results. One such example to illustrate is that dynamic box fusion and better bounding box methods refine candidate region proposals to achieve better Intersection over Union (IoU) values and better classification results, particularly with fine morphological changes in parasite egg classes [9]. These improvements are necessary in case of small-scale variations in parasitic eggs types.

   Other biomedical imaging problems that researchers have applied parasitic detection frameworks include helminth detection. In one of the studies customized YOLO variants trained on blood smear detecting malaria parasites were trained with the emphasis on a fast inference and low- computational complexity to permit clinical use [10]. This study can be used to detect parasitic eggs because the study is similar to the microscopic image analysis and the bounding box classification.

   Transfer learning-based models have demonstrated to perform well in recognizing the presence of parasites with the help of trained CNN backbones (ResNet and AlexNet). Such models are also more resistant to noise and changes in images as opposed to models that are trained on the ground [11]. This underscores the significance of model construction and domain fine-tuning to successful egg classification by its parasitic state.

   The general dynamics of the study of object detection suggest that the systems based on deep learning, including YOLO, Faster R-CNN, and DETR, outperform the traditional algorithms by a fairly significant margin. Hierarchical feature representations can be learned in these models and this increases their accuracy in automated microscopy tasks such as parasite eggs classification [12]. Changing feature engineering to data-driven learning has been one of the reasons behind the success of automated microscopy.

   According to the critical reviews of deep learning applications to agriculture and biomedical diagnostics, the usage of CNNs, transformers, and multi-task models is increasing, which justifies the prospect of these solutions in the context of parasitic egg recognition in the veterinary and clinical setting [13]. These surveys deal with common challenges like the unlabeled data, domain adaptation, and the issue of real-life implementation that directly affect parasitic diagnostic studies.

   Recent trends in research are solutions of Explainable AI (XAI) like Grad-CAM, which provides visual explanations to how a model has decided to operate. This is essential to the practitioners working in veterinary and medical environments because it allows them to make sense of model predictions and even minimize the error in the parasite classification task [14]. Explainability methods are used to facilitate transparency and trustworthiness in AI-based diagnostics.

   Specifically, edge-deployable systems such as those inspired by YOLO-like networks can be useful in real-time inference with resource constrained hardware. This is necessary when using it in field deployments in rural veterinary diagnostics where the computing resources are typically constrained [15]. YOLO-based models can be used

   in practice to detect objects and persons real-time in such environments, which is why they are suitable in veterinary practice.

3. PROPOSED SYSTEM

   The suggested system presents a deep learning model based on the YOLOv12 model of real-time detection and classification of parasitic eggs in fecal samples of goats. The system is capable of identifying the 11 different types of parasitic eggs using the sophisticated object detection features of YOLOv12, thus surpassing the manual microscopy drawbacks. The model is trained on a large set of parasitic eggs, and is specifically optimized to be used in a veterinary environment, which provides a scalable, efficient model to detect diseases at an early stage. The purpose of this system is to simplify veterinary diagnosis, better health management of animals, as well as, to increase disease control.

   Dataset Description:

   The training and test dataset used in the proposed system is obtained by using the IEEE DataPort competition in Parasitic Egg Detection and Classification using Microscopic image [1]. This is a collection of microscopic images of feces, labelled with 11 species of parasitic eggs. The training set is composed of 11,000 images being annotated 11, 031 times, and the testing set is composed of 2,200 images and annotated 2,228 times. Annotation was performed using YOLOv12 and the labels were arranged in the labels/train and labels/test folders. This was generated in a data.yaml file that would allow mapping the category id to the respective classes. Subclass mapping The following mapping is provided: Ascaris lumbricoides (ID: 0), Capillaria philippinensis (ID: 1), Enterobius vermicularis (ID: 2), Fasciolopsis buski (ID: 3), Hookworm egg (ID: 4), Hymenolepis nana (ID: 6) and Hymenolepis diminuta (ID: 5), Opisthorchis viverrine (ID: 7), Paragonimus spp (ID: 8), Taen Such a dataset is the basis of constructing a deep learning model that is able to detect and classify parasitic eggs in veterinary practices in real-time.

   Fig. 1. Working Flow.

   1. 1. YOLOV12

         The proposed deep learning-based object detector to detect parasitic eggs in the faeces of goats relies on the YOLOv12 that is an up-to-date one-stage model that is characterized by real-time functionality, speed, and accuracy. YOLO (You Only Look Once) is a powerful machine that recognizes objects and classifies them in one forward step, so it is suitable to use in the image recognition business. YOLOv12 is more accurate, more recalls and responds quicker than early models; it is able to process large and complex datasets. The model consists of a convoluted neural network (CNN) and splits input images into a grid of cells with each cell in the grid

         detecting an object in case its center falls within the cell. It predicts bounding boxes and the respective probabilities of object classes of multiple objects at the same time thus efficient object detection. The output of YOLOv12 consists of the coordinates of bounding boxes, confidence scores, and the probability distribution of the classes and is aimed at reducing the localization and classification errors.

         The loss in the YOLOv12 is divided into a few parts:

         1. Localization loss (for bounding box prediction),
         2. Confidence loss (for predicting objectness),
         3. Classification loss (for correct class identification). The total loss is determined by the following equation:

            = + +

            Where:

            • is a loss involving bounding box localization, often calculated by mean squared error (MSE).
            • is the loss in confidence scores, and it is a penalty against inaccurate identification of objects in a sample.
            • is the classification loss which is commonly computed based on the cross-entropy loss to the multi-class classification problem.

   The dataset used to train the model consists of parasitic egg images, with the use of YOLOv12 to distinguish and identify 11 parasitic egg species, such as: Ascaris lumbricoides, Hookworm, and Taenia spp. The pre-processed data is brought to a common input image size (e.g., 416×416 pixels) and random cropping, flipping, and color jittering are applied to the data to increase the overall generalization capacity of the model. YOLOv12 uses residual connectors and attention block, which enhance features extraction and enable the model to put emphasis on meaningful image features and limits the number of computations. Also, there is a Feature Pyramid Network (FPN) that is built to process multi-size objects, which is essential in the detection of eggs of different sizes. A batch size of 8 and 50 epochs are used to train the model to achieve optimum learning and convergence.

   The design comprises a number of phases:

   1. Input Layer: The input image is inputted to the network.
   2. Backbone: This component of the model derives high-level features on the image. YOLOv12 has simple and effective CNN backbones.
   3. Neck: Neck uses has pyramids of features and attention mechanisms that improve the features learned.
   4. Head: The head is used to predict bounding boxes, objectness and class scores of each grid cell.

   Fig. 2. Proposed Parasitic Egg Detection System.

   Pre-processing of the predictions is done by implementing non-maximum suppression (NMS) to eliminate the duplicate bounding box such that only the best-performing predictions are retained.

   The proposed parasitic egg detection system architecture (shown in Figure 2) works by taking fecal sample images and feeding the data into the YOLOv12 object detection model that processes the data to identify the different types of parasitic eggs.

   Overall, YOLOv12 offers an effective and precise system of the real-time identification of parasitic eggs in feces. The model is able to quickly handle large data sets because of its one stage architecture and optimization techniques and can easily be applied in veterinary diagnostics. Such a deep learning model will greatly enhance the process of identifying parasitic diseases thus allow early intervention and health management of the livestock.

4. RESULTS

   The results of the models such as YOLOv12 showed a high accuracy in detecting and classifying parasitic eggs, which were better than traditional methods with better accuracy, recall, and real-time inference in veterinary diagnostics.

   TABLE 1: PERFORMANCE OF CAPRIPARASITE AI SYSTEM FOR PARASITIC EGG DETECTION

   Class	Precision (P)	Recall (R)	mAP50	mAP50- 95
   Ascaris lumbricoides	0.961	0.967	0.985	0.866
   Capillaria philippinensis	0.961	0.944	0.991	0.804
   Enterobius vermicularis	0.981	0.968	0.974	0.903
   Fasciolopsis buski	0.953	0.988	0.990	0.993
   Hookworm egg	0.963	0.977	0.773	0.993
   Hymenolepis diminuta	0.951	0.988	0.995	0.964
   Hymenolepis nana	0.968	0.961	0.994	0.951
   Opisthorchis viverrine	0.948	0.990	0.989	0.956
   Paragonimus spp.	0.951	0.956	0.976	0.956
   Taenia spp. egg	0.956	0.970	0.997	0.916
   Trichuris trichiura	0.978	1	0.995	0.995

   In the CapriParasite AI, which is an AI implementation of YOLOv12, the results showed good detection performance when identifying caprine-based parasitic species. Other important statistics, such as a mean Average Precision (mAP50) of 0.995 and mAP50-95 of 0.905 show that it is a highly detecting metric. The highest mAP50- 95 value of 0.868 was recorded by Ascaris lumbricoides, with the other species such as Capillaria philippinensis and

   Hymenolepis nana recording 0.804 and 0.815 as score, respectively. The inference rate of the system is 25.9ms, postprocessing is 0.2ms and preprocessing is 0.4ms on each image and is thus efficient in veterinary diagnostics in real time application. These findings show the feasibility of the system.

   The CapriParasite AI system, based on the use of YOLOv12, has a confusion array that demonstrates its work in identifying 11 types of parasites, as ell as unspecified. The diagonal values, like 82 on Ascaris lumbricoides, and 104 on Paragonimus spp., are the right predictions which testify to the good accuracy of the model in the most species. Non-diagonal data, such as the 7 instances of Ascaris falsely classified as background, however, show resources of improvement. It is interesting to note that species such as Hymenolepis nana (91) and Trichuris trichiura (92) are very precise whereas Taenia spp. egg (67) has a higher misclassification. In general, the matrix depicts a strong system of detection, with little confusion among species, which proves its capability in providing an accurate and efficient veterinary diagnosis. Figure 3 represents the confusion matrix, which is a visual representation of the model performances.

   Fig. 3. Confusion Matrix.

   Fig. 4. F1-Confidence Curve

   The F1-Confidence Curve of CapriParasite AI which uses YOLOv12 shows the correlation between F1 score and threshold of confidence with 11 parasite species, and an average. The curve maxes out at an F1 score of 0.96 when the confidence level is 0.646, which indicates a good compromise between precision and recall. Other species such as Ascaris lumbricoides and Trichuris trichiura yield good results and their curves reflect values that are always near 0.9. Contrary,

   other species such as Taenia spp. egg have a lower curve meaning they have moderate detection performance. Figure 4 shows the F1- Confidence curve of the system, which shows robust and flexible detection of the system.

   Fig. 5. Precision-Confidence Curve

   The Precision-Confidence Curve of CapriParasite AI using YOLOv12 shows the performance of precision of 11 parasitic species, and a general average. The curve shows its peak at a precise of 1.00 at a confidence level of 1.000 indicating that the accuracy is almost near-perfect when confidence level is maximized. The parasite species Ascaris lumbricoides and Trichloroform Trichurus trichiura have a high consistency of 0.9 or higher over a very wide range of confidence values. Nonetheless, other species such as Capillaria philippinensis exhibit slight decreases in accuracy, and this means that there is a degree of variability in detection. The reliability of this system is highlighted in this graph and well-tuned confidence thresholds guarantee high accuracy and useful parasitic detection of the system in veterinary use. Figure 5 shows the Precision-Confidence curve, which is a graphical display of the precision of the model when using various levels of confidence.

   Fig. 6. Precision-Recall Curve

   The Precision-Recall Curve of CapriParasite AI, which is driven by YOLOv12, evaluates the performance of the model based on the performance of 11 parasitic species, as well as a general average. The system obtains a high mean Average Precision (mAP@0.5) of 0.985 and high precision is found over a large recall range. Exceptional species such as Ascaris lumbricoides (0.992) and Fasciolopsis buski (0.995) are highly precise which proves that the model is highly accurate in the detection of the two parasites. Nonetheless, the species like Capillaria philippinensis (0.964) and Opisthorchis viverrine (0.969) have a slightly lower precision. The curve represents an excellent trade-off between precision and recall, with a focus on the model to provide effective and reliable parasites detection as used in veterinary diagnostics. Figure 6 presents the Precision-Recall curve, which is the visual

   representation of the model performance depending on the classes of parasites.

   Fig. 7. Recall-Confidence Curve

   The Recall-Confidence Curve of CapriParasite AI, which is based on YOLOv12, measures the recall performance of 11 parasitic species, as well as the average one. With a confidence threshold of 0.000, all classes have a perfect recall of 1.00 as indicated, indicating that the model can be used to identify all cases, irrespective of the confidence index level. The five species, including Ascaris lumbricoides and Fasciolopsis buski, have a high recall within a large confidence threshold range of 1.0. The recall declines slowly with confidence, species such as Opisthorchis viverrine having some variation. This is to imply that it is very effective at detecting most of the cases at low thresholds and thus the system is very efficient in complete parasite detection in veterinary diagnostics. Figure 7 represents the Recall- Confidence curve, which demonstrates the ability of the model at different confidence levels.

5. CONCLUSION

CapriParasite AI system is an outstanding system that can identify and categorize parasitic egg in caprine feces using the YOLOv12 Power. The findings point to the strong detection features of the system, high precision, recall, and mAP scores on various parasitic species. The system is always better in accuracy and efficiency compared to the traditional diagnostic procedures thus, it offers a good solution in the field of veterinary diagnostics. The performance indicators, such as precision-recall curves, F1 confidence curves and confusion matrices demonstrate the capability of the system to discriminate between parasitic eggs with low error rates and false positives, which is essential to receive timely veterinary care.

With the application of the deep learning method and specifically YOLOv12, the system does not only have high detection accuracy but also matches in processing speed. CapriParasite AI is best suited to real-time analysis with preprocessing (0.4ms) and inference (25.9ms) and postprocessing (0.2ms) times per image, which makes it effective to implement in the real world (veterinary clinics and farms). Such efficiency can enable the screening of large datasets very fast, so that the possible infections may be determined in a short period of time and in an accurate way, which will result in improved management and control of the parasitic diseases in livestock.

Although the CapriParasite AI system that is currently in use has been showing a high level of performance, there are a

few aspects that could be improved. Future research may concentrate on the development of the model to identify a larger variety of parasitic species to make it more adaptable and applicable to different animal populations. The use of other sources of data, including environmental conditions or geographic data, may enhance the quality of diagnostic further because it offers some contextual information on the prevalence of parasites. More so, the use of explainable AI (XAI) methods may assist veterinarians to have a clearer understanding of the predictions of the model, thus rendering it more understandable and reliable. Lastly, to enhance the functionality of the system and improve its efficiency in processing higher-resolution images and dealing with real- world variability including lighting conditions and sample contaminations would make the system robust and reliable in a variety of environments.

REFERENCES

1. N. Anantrasirichai et al., ICIP 2022 CHALLENGE ON PARASITIC EGG DETECTION AND CLASSIFICATION IN MICROSCOPIC IMAGES: DATASET, METHODS AND RESULTS, Proceedings –

   International Conference on Image Processing, ICIP, pp. 43064310, 2022, doi: 10.1109/ICIP46576.2022.9897267.

2. Parasitic Egg Detection and Classification with Transformer-Based Architectures | Request PDF. Accessed: Feb. 03, 2026. [Online]. Available: https://www.researchgate.net/publication/365108939_Parasitic_Egg_D etection_and_Classification_with_Transformer-Based_Architectures
3. Y. Wang, Z. He, S. Huang, and H. Du, A ROBUST ENSEMBLE

   MODEL FOR PARASITIC EGG DETECTION AND

   CLASSIFICATION, Proceedings – International Conference on Image Processing, ICIP, pp. 42584262, 2022, doi: 10.1109/ICIP46576.2022.9897192.

4. W. Xu, Q. Zhai, J. Liu, X. Xu, and J. Hua, A lightweight deep-learning model for parasite egg detection in micoscopy images, Parasites and Vectors , vol. 17, no. 1, pp. 454-, Dec. 2024, doi: 10.1186/S13071-024- 06503-2/TABLES/6.
5. Z. Wan, S. Liu, F. Ding, M. Li, G. Srivastava, and K. Yu, C2BNet: A Deep Learning Architecture With Coupled Composite Backbone for Parasitic Egg Detection in Microscopic Images, IEEE J. Biomed. Health Inform., vol. 29, no. 4, pp. 23532364, 2025, doi: 10.1109/JBHI.2023.3318604.
6. Z. Wan, S. Liu, and M. Li, Dynamic boxes fusion strategy in object detection, Jul. 2022, Accessed: Feb. 03, 2026. [Online]. Available: https://arxiv.org/abs/2207.00997v3
7. J. Redmon and A. Farhadi, YOLOv3: An Incremental Improvement, Apr. 2018, Accessed: Feb. 03, 2026. [Online]. Available: https://arxiv.org/abs/1804.02767v1
8. L. Jiao et al., A survey of deep learning-based object detection, IEEE Access, vol. 7, pp. 128837128868, 2019, doi: 10.1109/ACCESS.2019.2939201.
9. N. Aldahoul, H. A. Karim, S. Limson Kee, and M. J. Toledo Tan, LOCALIZATION AND CLASSIFICATION OF PARASITIC EGGS IN MICROSCPIC IMAGES USING AN EFFICIENTDET DETECTOR.
10. (PDF) Deep Learning for Real-time Malaria Parasite Detection and Counting Using YOLO-mp. Accessed: Feb. 03, 2026. [Online]. Available: https://www.researchgate.net/publication/363732534_Deep_Learning_ for_Real- time_Malaria_Parasite_Detection_and_Counting_Using_YOLO- mp?utm_source=chatgpt.com
11. S. Kumar, T. Arif, G. Ahamad, A. A. Chaudhary, S. Khan, and M. A.

    M. Ali, An Efficient and Effective Framework for Intestinal Parasite Egg Detection Using YOLOv5, Diagnostics 2023, Vol. 13, vol. 13, no. 18, Sep. 2023, doi: 10.3390/DIAGNOSTICS13182978.

12. N. Penpong, Y. Wanna, C. Kamjanlard, A. Techasen, and T. Intharah, Attacking the out-of-domain problem of a parasite egg detection in-the- wild, Heliyon, vol. 10, no. 4, p. e26153, Feb. 2024, doi: 10.1016/J.HELIYON.2024.E26153.
13. N. AlDahoul, H. A. Karim, M. A. Momo, F. I. F. Escobar, V. A.

    Magallanes, and M. J. T. Tan, Parasitic egg recognition using

    convolution and attention network, Scientific Reports 2023 13:1, vol. 13, no. 1, pp. 14475-, Sep. 2023, doi: 10.1038/s41598-023-41711-3.

14. T. Suwannaphong, S. Chavana, S. Tongsom, D. Palasuwan, T. H. Chalidabhongse, and N. Anantrasirichai, Parasitic Egg Detection and Classification in Low-cost Microscopic Images using Transfer Learning, SN Comput. Sci., vol. 5, no. 1, Jul. 2021, doi: 10.1007/s42979-023-02406-8.
15. Z. H. Aung, K. Srithaworn, and T. Achakulvisut, Multitask learning via pseudo-label generation and ensemble prediction for parasitic egg cell detection: IEEE ICIP Challenge 2022, Proceedings – International Conference on Image Processing, ICIP, pp. 42734277, 2022, doi: 10.1109/ICIP46576.2022.9897464.