R-CNN Research Articles

Using a Region-Based Convolutional Neural Network (R-CNN) for Potato Segmentation in a Sorting Process

This study focuses on the segmentation part in the development of a potato-sorting system that utilizes camera input for the segmentation and classification of potatoes. The key challenge addressed is the need for efficient segmentation to allow the sorter to handle a higher volume of potatoes simultaneously. To achieve this, the study employs a region-based convolutional neural network (R-CNN) approach for the segmentation task, while trying to achieve more precise segmentation than with classic CNN-based object detectors. Specifically, Mask R-CNN is implemented and evaluated based on its performance with different parameters in order to achieve the best segmentation results. The implementation and methodologies used are thoroughly detailed in this work. The findings reveal that Mask R-CNN models can be utilized in the production process of potato sorting and can improve the process.

Optimization of Decision Support Technology for Offshore Oil Condition Monitoring with Carbon Neutrality as the Goal in the Enterprise Development Process.

This study aims to explore the integration of the Faster R-CNN (Region-based Convolutional Neural Network) algorithm from deep learning into the MobileNet v2 architecture, within the context of enterprises aiming for carbon neutrality in their development process. The experiment develops a marine oil condition monitoring and classification model based on the fusion of MobileNet v2 and Faster R-CNN algorithms. This model utilizes the MobileNet v2 network to extract rich feature information from input images and combines the Faster R-CNN algorithm to rapidly and accurately generate candidate regions for oil condition monitoring, followed by detailed feature fusion and classification of these regions. The performance of the model is evaluated through experimental assessments. The results demonstrate that the average loss value of the proposed model is approximately 0.45. Moreover, the recognition accuracy of the model for oil condition on the training and testing sets reaches 90.51% and 93.08%, respectively, while the accuracy of other algorithms remains below 90%. Thus, the model constructed in this study exhibits excellent performance in terms of loss value and recognition accuracy, providing reliable technical support for offshore oil monitoring and contributing to the promotion of sustainable utilization and conservation of marine resources.

Bridging technology and ecology: enhancing applicability of deep learning and UAV-based flower recognition

The decline of insect biomass, including pollinators, represents a significant ecological challenge, impacting both biodiversity and ecosystems. Effective monitoring of pollinator habitats, especially floral resources, is essential for addressing this issue. This study connects drone and deep learning technologies to their practical application in ecological research. It focuses on simplifying the application of these technologies. Updating an object detection toolbox to TensorFlow (TF) 2 enhanced performance and ensured compatibility with newer software packages, facilitating access to multiple object recognition models - Faster Region-based Convolutional Neural Network (Faster R-CNN), Single-Shot-Detector (SSD), and EfficientDet. The three object detection models were tested on two datasets of UAV images of flower-rich grasslands, to evaluate their application potential in practice. A practical guide for biologists to apply flower recognition to Unmanned Aerial Vehicle (UAV) imagery is also provided. The results showed that Faster RCNN had the best overall performance with a precision of 89.9% and a recall of 89%, followed by EfficientDet, which excelled in recall but at a lower precision. Notably, EfficientDet demonstrated the lowest model complexity, making it a suitable choice for applications requiring a balance between efficiency and detection performance. Challenges remain, such as detecting flowers in dense vegetation and accounting for environmental variability.

Impact of Skin Lesion Descriptors and Deep Learning Architecture for the Effective Detection and Identification of Skin Diseases - A Systematic Review

In diagnosing skin diseases, segmentation of lesion region and classification of detected lesion type are the two major processes. This paper conducts a systematic review related to the lesion descriptors extracted from the skin region and the machine learning classifiers utilized in differentiating the descriptor types and also discusses the deep learning schemes that impact skin lesion diagnosis. More specifically, the paper focuses on lesion region detection-based deep learning schemes that are derived from the Mask Region-based CNN (RCNN), U-Net, and DeepLabV3+ architectures. In the case of skin lesion segmentation, the recent multi-attention scheme results in a recall, precision, and F1-score of 93.97%, 93.94%, and 93.73% respectively in the ISIC-2016 dataset which is higher than other lesion detection approaches. In the case of skin lesion classification, the Diverse CNN approach results in a maximum accuracy, recall, and precision of 96.12%, 93.11%, and 94.63% respectively when evaluated using the HAM-10000 dataset which is higher than other lesion classification approaches.

Innovative Human Interaction System to Predict College Student Emotions Using the Extended MASK-R-CNN Algorithm

There is a rising demand for emerging machines that can be self-decisive and intelligent. Machines can capture the emotions and gestures of college students to mechanise tasks and handle interactions better. Facial expressions based on emotion recognition are practices that play a substantial role in the modern fields of artificial intelligence and computer vision. Numerous manual methods for detecting emotions are focused on few basic emotions. Additionally, significant time is needed for appropriate detection. Nonetheless, these techniques are time-consuming and inefficient for obtaining better results. Therefore, an effective object detection model is needed to address such issues. To overcome these challenges, several studies have focused on object detection systems to provide effective emotion prediction. Conversely, it results in a lack of speed, precision and computational complexity. To improve object detection performance, the proposed model employs deep learning (DL)-based adaptive feature spatial anchor refinement with a mask region-based convolutional neural network (Mask RCNN). It uses the Facial Expression Recognition (FER) 2013 dataset for the evaluation process. Correspondingly, the efficacy of the projected model is calculated via various evaluation metrics, such as the recall, precision and mean average precision (mAP), to estimate the performance of the proposed DL method. It achieves 0.75298 for MAP@50, 0.70252 for precision and 0.66606 for recall. Furthermore, a comparison of existing models reveals the efficiency of the proposed DL method. The present research is intended to contribute to emerging object detection methods for enhancing real-time analysis of student emotions in various environments, such as classrooms and online education.

Artificial Intelligence-Driven Optimization of Carbon Neutrality Strategies in Population Studies

With the growing severity of global climate change, achieving carbon neutrality has become a central focus worldwide. The intersection of population studies and carbon neutrality introduces significant challenges in predicting and optimizing energy consumption, as demographic factors play a crucial role in shaping carbon emissions. This paper proposes a model based on a Region-based Convolutional Neural Network (RCNN) and Generative Adversarial Network (GAN), enhanced with a dual-stage attention mechanism for optimization. The model automatically extracts key features from complex demographic and carbon emission data, leveraging the attention mechanism to assign appropriate weights, thereby capturing the behavioral patterns and trends in energy consumption driven by population dynamics more effectively. By integrating multi-source data, including historical carbon emissions, population density, demographic trends, meteorological data, and economic indicators, experimental results demonstrate the model's outstanding performance across multiple datasets.

ML-UrineQuant: A machine learning program for identifying and quantifying mouse urine on absorbent paper.

The void spot assay has gained popularity as a way of assessing functional bladder voiding parameters in mice, but analyzing the size and distribution of urine spot patterns on filter paper with software remains problematic due to inter-laboratory differences in image contrast and resolution quality and non-void artifacts. We have developed a machine learning algorithm based on Region-based Convolutional Neural Networks (Mask-RCNN) that was trained in object recognition to detect and quantitate urine spots across a broad range of sizes-ML-UrineQuant. The model proved extremely accurate at identifying urine spots in a wide variety of illumination and contrast settings. The overwhelming advantage it offers over current algorithms will be to allow individual labs to fine-tune the model on their specific images regardless of the image characteristics. This should be a valuable tool for anyone performing lower urinary tract research using mouse models.

Fusion of hyperspectral imaging and object detection for early disease diagnosis in resource-limited healthcare systems

The fusion of hyperspectral imaging (HSI) and advanced object detection techniques holds transformative potential for early disease diagnosis, particularly in resource-limited healthcare systems. Hyperspectral imaging, which captures detailed spectral information across numerous wavelength bands, enables the detection of subtle physiological changes that are often imperceptible in conventional imaging methods. This non-invasive imaging modality provides comprehensive insights into tissue composition, facilitating the early identification of diseases such as cancer, diabetic retinopathy, and skin disorders. However, the high-dimensional nature of HSI data presents challenges in processing and analysis, necessitating the integration of sophisticated object detection algorithms. Object detection, powered by machine learning and deep learning models, enhances the capability to identify and classify pathological features within hyperspectral datasets with high precision and efficiency. Techniques such as convolutional neural networks (CNNs) and region-based convolutional neural networks (R-CNNs) have proven effective in extracting critical features and localizing disease-specific patterns in HSI data. The fusion of these technologies not only improves diagnostic accuracy but also optimizes computational resources, making them suitable for deployment in healthcare systems with limited infrastructure. In resource-constrained environments, where access to advanced diagnostic tools is limited, the combined application of HSI and object detection can bridge critical gaps. By enabling rapid, accurate, and cost-effective disease screening, this approach enhances early diagnosis and improves patient outcomes. This study explores the methodologies, applications, and potential challenges of integrating hyperspectral imaging with object detection, emphasizing its role in advancing healthcare delivery in under-resourced settings.

Image Augmentation Approaches for Building Dimension Estimation in Street View Images Using Object Detection and Instance Segmentation Based on Deep Learning

There are numerous applications for building dimension data, including building performance simulation and urban heat island investigations. In this context, object detection and instance segmentation methods—based on deep learning—are often used with Street View Images (SVIs) to estimate building dimensions. However, these methods typically depend on large and diverse datasets. Image augmentation can artificially boost dataset diversity, yet its role in building dimension estimation from SVIs remains under-studied. This research presents a methodology that applies eight distinct augmentation techniques—brightness, contrast, perspective, rotation, scale, shearing, translation augmentation, and a combined “sum of all” approach—to train models in two tasks: object detection with Faster Region-Based Convolutional Neural Networks (Faster R-CNNs) and instance segmentation with You Only Look Once (YOLO)v10. Comparing the performance with and without augmentation revealed that contrast augmentation consistently provided the greatest improvement in both bounding-box detection and instance segmentation. Using all augmentations at once rarely outperformed the single most effective method, and sometimes degraded the accuracy; shearing augmentation ranked as the second-best approach. Notably, the validation and test findings were closely aligned. These results, alongside the potential applications and the method’s current limitations, underscore the importance of carefully selected augmentations for reliable building dimension estimation.

Segmentation and severity classification of scar tissues in LGE-CMR images using HDResC-Net with Flamingo gannet search optimization.

Late gadolinium enhanced-cardiac magnetic resonance (LGE-CMR) images play a critical role in evaluating cardiac pathology, where scar tissue serves as a vital indicator impacting prognosis and treatment decisions. However, accurately segmenting scar tissues and assessing their severity present challenges due to complex tissue composition and imaging artifacts. Existing methods often lack precision and robustness, limiting their clinical applicability. This work proposes a novel methodology that integrates the optimal segmentation algorithm (OSA) for segmentation and Flamingo Gannet search optimization-enabled hybrid deep residual convolutional network (FGSO-HDResC-Net) for severity classification of scar tissues in LGE-CMR images. Initially, the input image is pre-processed by using the adaptive Gabor Kuwahara filter. Then, the approach combines myocardium segmentation via region-based convolutional neural network and scar segmentation using OSA. Subsequently, FGSO-HDResC-Net integrates feature extraction and classification while optimizing hyperparameters through Flamingo Gannet search optimization. The feature extraction stage introduces two sets of techniques: localization features with texture analysis and spatial/temporal features using a deep residual network, complemented by feature fusion using the fractional concept. These features are inputted into a customized 1D convolutional neural network model for severity classification. Through comprehensive evaluation, the effectiveness of FGSO-HDResC-Net in accurately classifying scar tissue severity is demonstrated, offering improved disease assessment and treatment planning for cardiac patients. Moreover, the proposed FGSO-HDResC-Net model demonstrated superior performance, achieving an accuracy of 96.45%, a true positive rate of 95.42%, a true negative rate of 96.48%, a positive predictive value of 94.20%, and a negative predictive value of 94.18%. The accuracy of the devised model is 14.50%, 12.99%, 10.74%, 9.75%, 12.79%, and 11.26% improved than the traditional models.

Some artificial intelligence tools may currently be useful in orthopedic surgery and traumatology.

Artificial intelligence (AI) can help in diagnosing fractures and demonstrating effusions, dislocations, and focal bone lesions in both adult and pediatric aged individuals and also aid in early tumor discovery (bone osteosarcoma) and in robot-assisted surgery. A recent AI model [Mask R-CNN (region-based convolutional neural network)] has shown to be dependable for detecting surgical target zones in pediatric hip and periarticular infections, offering a more convenient and quicker alternative to conventional methods. It can help inexperienced physicians in pre-treatment evaluations, diminishing the risk of missed diagnosis and misdiagnosis. AI has some very interesting applications in orthopedic surgery, which orthopedic surgeons should be aware of and if possible use. Although some interesting advances have been made recently on AI in orthopedic surgery, its usefulness in clinical practice is still very limited. Ethical concerns, such as transparency in AI decision-making, data privacy, and the potential loss of human intuition cannot be forgotten. Besides, it is paramount to explore how to gain trust from both healthcare professionals and patients in the utilization of AI.

Development of a Deep Mask Region-Based Convolutional Neural Network with Improved Weighted Quantum Wolf Optimization for Sickle Cell Anaemia Detection and Classification

To effectively manage Sickle Cell Anaemia (SCA), a severe inherited blood condition, early detection and accurate categorization are essential. Timely and precise identification is hindered by the high error rates, limited scalability, and dependency on personal expertise associated with conventional diagnostic approaches. To address these challenges, this study proposes a novel Deep Mask Region-Based Convolutional Neural Network (DMRCNN) combined with Improved Weighted Quantum Wolf Optimization (IWQWO) for SCA identification and classification. By leveraging advanced feature extraction and segmentation methods, the DMRCNN enables precise localization and identification of abnormal cells in blood smear images. The IWQWO method optimizes the DMRCNN's hyper parameters enhancing the network's efficiency by achieving an optimal balance between convergence speed and prediction accuracy. The primary objectives of this study are to improve classification accuracy, reduce computational overhead, and minimize false positives and false negatives in SCA diagnostics. Experimental results demonstrate that the proposed system achieves 96.8% precision, 97.2% accuracy, 96.5% recall and 96.8% F1-score, surpassing existing approaches with superior metrics. These findings underscore the effectiveness of the proposed method in automating SCA detection and classification, paving the way for potential integration into clinical workflows for accurate and timely assessments.

Leveraging deep semantic segmentation for assisted weed detection

In agriculture, it is crucial to identify and control weeds as these plant species pose a significant threat to the growth and development of crops by competing for vital resources such as nutrients, water, and light. A promising solution to this problem is adopting smart weed control systems (SWCS) that significantly reduce the use of harmful chemical products. Furthermore, SWCS leads to reduced production costs and a more sustainable and eco-friendly approach to farming. However, implementing SWCS in natural fields can be challenging, mainly due to difficulties in accurately localizing plants. To address this issue, a visual identification system can be employed to label plants from images using a process known as semantic segmentation. In this work, we have implemented, validated, and compared three deep learning approaches, including Mask Region-based Convolutional Neural Network (Mask R-CNN), Mask R-CNN enhanced with an Atrous Spatial Pyramid Pooling module (Mask R-CNN-ASPP), and a proposed model named Residual U-Net architecture, for the semantic pixel segmentation of high densities of both crops (Zea mays) and weeds (including narrow-leaf weeds and broad-leaf weeds). Data augmentation and transfer learning have also been implemented. The performance of the models was evaluated with the well-known metrics Precision, Recall, Dice similarity coefficient (DSC), and mean Intersection-Over-Union (mIoU). As a result of the analysis, the DSC and mIoU of Mask R-CNN-ASPP based models were up to 10.63% and 10.54% superior to that of the Mask R-CNN based models. Nonetheless, the proposed Residual U-Net architecture outperformed Mask R-CNN-ASPP based networks in all the metrics, reaching a DSC of 92.98% and mIoU of 87.12%. Thus, we have concluded that the proposed Residual U-Net-like architecture is the best alternative for the semantic segmentation task in images with high plant density. Our research addresses the challenge of weed identification and control in agriculture, helping farmers produce crops more efficiently while minimizing environmental impact.

Hybrid Model Compression Pipeline for Efficient Object Detection using Parallel Recurrent Convolutional Neural Networks

Several Models of Deep Learning (DL) have demonstrated impressive performance across multiple object detection problems. Large object detection methods based on DL are typically computationally and memory-intensive. Hence, this paper presents model compression strategies for object identification with Parallel Recurrent Convolutional Neural Networks (MCS-OD-PRCNN). Initially, the input photos come from the Common Objects in Context (COCO) 2017 dataset. Next, using Improved Bilateral Texture Filtering (IBTF), the input images are pre-processed. The pre-processed images are then given to the suggested deep-learning model, Parallel Recurrent Convolutional Neural Networks (PRCNNs), which identifies and localizes the objects in the image. After training and validating the PRCNN model on the pre-processed dataset, compression model precision is decreased with the application of strategies like quantization and pruning, eliminating redundant weights and connections, and training a smaller, more efficient student model based on the larger PRCNN model. To ensure optimal performance, hybrid fox and chimp optimization algorithms (Hyb-FCOA) are employed for the compression model’s parameter tuning. The suggested methodology is carried out in Python environment, and fundamental evaluation metrics such as Accuracy, Precision, Recall, F-Measure, mean Average Precision (mAP), Matthew’s Correlation Coefficient (MCC), Intersection over Union (IoU), and Positive Predictive Value (PPV) are employed to evaluate the strategy's performance. The proposed method attains 20.08%, 23.35%, and 27.79% higher accuracy compared to existing techniques such as using one-to-one instruction and guided hybrid quantization for remote sensing object detection (GHOST-GQSD), Fast Region-Based Convolutional Neural Network (Fast-RCNN), and You Only Look Once version 4 (YOLOv4), respectively.

Detection of oscillation-like patterns in eclipsing binary light curves using neural network-based object detection algorithms

The primary aim of this research is to evaluate several convolutional neural network-based object detection algorithms for identifying oscillation-like patterns in light curves of eclipsing binaries. This involved creating a robust detection framework that can effectively process both synthetic light curves and real observational data. The study employs several state-of-the-art object detection algorithms, including Single Shot MultiBox Detector, Faster Region-based Convolutional Neural Network, You Only Look Once, and EfficientDet, as well as a custom non-pretrained model implemented from scratch. Synthetic light curve images and images derived from observational TESS light curves of known eclipsing binaries with a pulsating component were constructed with corresponding annotation files using custom scripts. The models were trained and validated on established datasets, which was followed by testing on unseen Kepler data to assess their generalisation performance. The statistical metrics were also calculated to review the quality of each model. The results indicate that the pre-trained models exhibit high accuracy and reliability in detecting the targeted patterns. The Faster Region-based Convolutional Neural Network and You Only Look Once in particular showed superior performance in terms of object detection evaluation metrics on the validation dataset, including a mean average precision value exceeding 99%. The Single Shot MultiBox Detector, on the other hand, is the fastest, although it shows a slightly lower performance, with a mean average precision of 97%. These findings highlight the potential of these models to significantly contribute to the automated determination of pulsating components in eclipsing binary systems and thus facilitate more efficient and comprehensive astrophysical investigations.

Tomato plant disease prediction system with a new framework SSMAN using advanced deep learning techniques

Agriculture plays a pivotal role in India's economy, and the timely detection of plant infections is essential to safeguard crops and prevent further spread of diseases. The conventional approach involves manual inspection of plant leaves to identify the specific type of disease, a task typically carried out by farmers or plant pathologists. In previous studies, you only look once (YOLO) and faster region-based convolutional neural network (R-CNN), machine learning algorithms were applied to datasets for detecting objects on tomato leaves which includes a total of images 2403 and got accuracies of 86 and 82 percent. In this paper, a deep convolutional neural network (DCNN) model proposed with a new framework separate, shift, and merge based AlexNet50 algorithm (SSMAN) is used to predict the disease at an earlier stage with higher accuracy. Among various pre-trained deep models, AlexNet emerges as the top performer, achieving the highest accuracy in disease classification. SSMAN can address anomalies in images by employing a class decomposition approach to scrutinize class boundaries. AlexNet exhibits a notable accuracy of 98.30% in successfully identifying tomato leaf diseases from images, with pre-trained new framework, superior to the original AlexNet architecture as well as traditional classification methods with other algorithms.

Proximal femur segmentation and quantification in dual-energy subtraction tomosynthesis: A novel approach to fracture risk assessment.

Proximal femur segmentation and quantification in dual-energy subtraction tomosynthesis: A novel approach to fracture risk assessment.

Automated spinopelvic measurements on radiographs with artificial intelligence: a multi-reader study

PurposeTo develop an artificial intelligence (AI) algorithm for automated measurements of spinopelvic parameters on lateral radiographs and compare its performance to multiple experienced radiologists and surgeons.MethodsOn lateral full-spine radiographs of 295 consecutive patients, a two-staged region-based convolutional neural network (R-CNN) was trained to detect anatomical landmarks and calculate thoracic kyphosis (TK), lumbar lordosis (LL), sacral slope (SS), and sagittal vertical axis (SVA). Performance was evaluated on 65 radiographs not used for training, which were measured independently by 6 readers (3 radiologists, 3 surgeons), and the median per measurement was set as the reference standard. Intraclass correlation coefficient (ICC), mean absolute error (MAE), and standard deviation (SD) were used for statistical analysis; while, ANOVA was used to search for significant differences between the AI and human readers.ResultsAutomatic measurements (AI) showed excellent correlation with the reference standard, with all ICCs within the range of the readers (TK: 0.92 [AI] vs. 0.85–0.96 [readers]; LL: 0.95 vs. 0.87–0.98; SS: 0.93 vs. 0.89–0.98; SVA: 1.00 vs. 0.99–1.00; all p < 0.001). Analysis of the MAE (± SD) revealed comparable results to the six readers (TK: 3.71° (± 4.24) [AI] v.s 1.86–5.88° (± 3.48–6.17) [readers]; LL: 4.53° ± 4.68 vs. 2.21–5.34° (± 2.60–7.38); SS: 4.56° (± 6.10) vs. 2.20–4.76° (± 3.15–7.37); SVA: 2.44 mm (± 3.93) vs. 1.22–2.79 mm (± 2.42–7.11)); while, ANOVA confirmed no significant difference between the errors of the AI and any human reader (all p > 0.05). Human reading time was on average 139 s per case (range: 86–231 s).ConclusionOur AI algorithm provides spinopelvic measurements accurate within the variability of experienced readers, but with the potential to save time and increase reproducibility.

Cell-APP: A generalizable method for microscopic cell annotation, segmentation, and classification.

High throughput fluorescence microscopy is an essential tool in systems biological studies of eukaryotic cells. Its power can be fully realized when all cells in a field of view and the entire time series can be accurately localized and quantified. These tasks can be mapped to the common paradigm in computer vision: instance segmentation. Recently, supervised deep learning-based methods have become state-of-the-art for cellular instance segmentation. However, these methods require large amounts of high-quality training data. This requirement challenges our ability to train increasingly performant object detectors due to the limited availability of annotated training data, which is typically assembled via laborious hand annotation. Here, we present a generalizable method for generating large instance segmentation training datasets for tissue-culture cells in transmitted light microscopy images. We use datasets created by this method to train vision transformer (ViT) based Mask-RCNNs (Region-based Convolutional Neural Networks) that produce instance segmentations wherein cells are classified as "m-phase" (dividing) or "interphase" (non-dividing). While training these models, we also address the dataset class imbalance between m-phase and interphase cell annotations, which arises for biological reasons, using probabilistically weighted loss functions and partisan training data collection methods. We demonstrate the validity of these approaches by producing highly accurate object detectors that can serve as general tools for the segmentation and classification of morphologically diverse cells. Since the methodology depends only on generic cellular features, we hypothesize that it can be further generalized to most adherent tissue culture cell lines.

Deep-Learning-Assisted Digital Fluorescence Immunoassay on Magnetic Beads for Ultrasensitive Determination of Protein Biomarkers.

Digital fluorescence immunoassay (DFI) based on random dispersion magnetic beads (MBs) is one of the powerful methods for ultrasensitive determination of protein biomarkers. However, in the DFI, improving the limit of detection (LOD) is challenging since the ratio of signal-to-background and the speed of manual counting beads are low. Herein, we developed a deep-learning network (ATTBeadNet) by utilizing a new hybrid attention mechanism within a UNet3+ framework for accurately and fast counting the MBs and proposed a DFI using CdS quantum dots (QDs) with narrow peak and optical stability as reported at first time. The developed ATTBeadNet was applied to counting the MBs, resulting in the F1 score (95.91%) being higher than those of other methods (ImageJ, 68.33%; computer vision-based, 92.99%; fully convolutional network, 75.00%; mask region-based convolutional neural network, 70.34%). On principle-on-proof, a sandwich MB-based DFI was proposed, in which human interleukin-6 (IL-6) was taken as a model protein biomarker, while antibody-bound streptavidin-coated MBs were used as capture MBs and antibody-HRP-tyramide-functionalized CdS QDs were used as the binding reporter. When the developed ATTBeadNet was applied to the MB-based DFI of IL-6 (20 μL), the linear range from 5 to 100 fM and an LOD of 3.1 fM were achieved, which are better than those using the ImageJ method (linear range from 30 to 100 fM and LOD of 20 fM). This work demonstrates that the integration of the deep-learning network with DFI is a promising strategy for the highly sensitive and accurate determination of protein biomarkers.