Feature Map Research Articles

A novel deep learning framework for retinal disease detection leveraging contextual and local features cues from retinal images.

Retinal diseases are a serious global threat to human vision, and early identification is essential for effective prevention and treatment. However, current diagnostic methods rely on manual analysis of fundus images, which heavily depends on the expertise of ophthalmologists. This manual process is time-consuming and labor-intensive and can sometimes lead to missed diagnoses. With advancements in computer vision technology, several automated models have been proposed to improve diagnostic accuracy for retinal diseases and medical imaging in general. However, these methods face challenges in accurately detecting specific diseases within images due to inherent issues associated with fundus images, including inter-class similarities, intra-class variations, limited local information, insufficient contextual understanding, and class imbalances within datasets. To address these challenges, we propose a novel deep learning framework for accurate retinal disease classification. This framework is designed to achieve high accuracy in identifying various retinal diseases while overcoming inherent challenges associated with fundus images. Generally, the framework consists of three main modules. The first module is Densely Connected Multidilated Convolution Neural Network (DCM-CNN) that extracts global contextual information by effectively integrating novel Casual Dilated Dense Convolutional Blocks (CDDCBs). The second module of the framework, namely, Local-Patch-based Convolution Neural Network (LP-CNN), utilizes Class Activation Map (CAM) (obtained from DCM-CNN) to extract local and fine-grained information. To identify the correct class and minimize the error, a synergic network is utilized that takes the feature maps of both DCM-CNN and LP-CNN and connects both maps in a fully connected fashion to identify the correct class and minimize the errors. The framework is evaluated through a comprehensive set of experiments, both quantitatively and qualitatively, using two publicly available benchmark datasets: RFMiD and ODIR-5K. Our experimental results demonstrate the effectiveness of the proposed framework and achieves higher performance on RFMiD and ODIR-5K datasets compared to reference methods.

Optimized Watermelon Scion Leaf Segmentation Model Based on Hungarian Algorithm and Information Theory

In the fully automated grafting process of watermelon seedlings, it is crucial to ensure that the scion’s cotyledons maintain a perpendicular orientation with the rootstock cotyledons. To achieve precise segmentation of watermelon scion cotyledons and accurately extract parameters, such as cotyledon orientation angles, this study introduces enhancements to the Mask2Former network, aiming to improve segmentation accuracy for watermelon scion cotyledons. Specifically, two innovative modules are designed. Taking Swin-Former as the backbone, an Optimal Feature Re-ranking (OFR) module based on the Hungarian Algorithm is devised to re-rank the feature maps obtained from the feature extraction process. Grounded in information theory, the amount of information in semantic segmentation tasks is quantified as Shannon entropy, enabling the model to perceive the information distribution of the feature maps and dynamically adjust the output features. Experimental results demonstrate that the improved model achieves mIoU, mDice, mPrecision, and mRecall scores of 97.44%, 98.70%, 98.20%, and 99.21%, respectively, greatly outperforming Mask2Former, FCNN, and DeepLabv3. Furthermore, the enhanced network exhibits superior accuracy in low signal-to-noise ratio environments, highlighting its robustness in complex scenarios. This study provides a high-precision solution for agricultural automation in the watermelon industry, contributing to the development of fully automated grafting machines.

Infrared dim tiny-sized target detection based on feature fusion

Detection of infrared objects is essential for applications ranging from remote sensing to thermal imaging. In certain instances, such as when the infrared target is situated at a considerable distance from the detector, the detected object exhibits diminutive dimensions with a concurrently low signal intensity, leading to a challenge in achieving precision in object detection. In this paper, we propose a dual-layer omnidirectional target enhancement (DODTE) module to address the issue of low contrast, which causes the target to be submerged in the background and lose its position information. The module aims to extract and enhance the position information of the target, serving as a feature map for subsequent guidance. Furthermore, to tackle the problem of the fuzzy and ambiguous shape of the target caused by a weak signal and tiny dimension, a residual-based pyramid-like (RBPL) module is designed, which extracts the deep information (i.e. the shape information) from the images to compensate for the lack of expressive ability of the fixed convolution kernel for shape information. These two main modules are employed as the core to form a feature fusion network to realize the detection of infrared dim tiny-sized targets. The comparison of the proposed network with other algorithms are performed on open-source dataset and experimentally generated infrared images. The quantitative evaluation metrics, including IOUs and F1 scores, validate the outperformance of the proposed network. Furthermore, ablation experiments demonstrate that the proposed two modules can effectively tackle the tiny-size, low contrast and dark intensity issues, providing a solution for detecting dim and small-size infrared targets.

Development of a Lightweight Cross-Scale Decoupling Feature Fusion Network for Surface Defect Detection in Permanent Magnets

Abstract Permanent magnet motors may suffer from imperceptible localized demagnetization due to surface damage during operation. This paper proposes a lightweight cross-scale decoupling feature fusion network (LCDFFN) for permanent magnets defect detection, which introduces separable convolution in the backbone and incorporates the Adaptive Multi-scale Feature Fusion with Interactive Attention Mechanism (AMFIA) to enhance small target detection. The Adaptive Attention Mechanism (AAM) and Feature Enhancement Module (FEM) based on dilated convolution are introduced in the feature fusion network, improving multi-scale object detection. Finally, the decoupled feature prediction network outputs the defect identification feature map. In addition, we also propose a novel fuzzy intersection over the union loss function. LCDFFN achieves a mAP@0.5 of 97.8%, with 108.4M parameters and a detection speed of 113.62 FPS. The proposed method significantly improves defect image detection for permanent magnets and is highly practical for industrial production.

Prenatal Diagnostics Using Deep Learning: A Dual Approach to Plane Localization and Cerebellum Segmentation in Ultrasound Images.

The fetal ultrasound examination is the significant task of mid-term pregnancy inspection and the accurate localization as well as the segmentation of the cerebellum is crucial for clinical diagnosis. This research focuses on developing deep learning techniques for prenatal prediction of neurodevelopmental disorders using 5th-month ultrasound brain images. The study introduces two specialized convolutional neural network (CNN) architectures: the differential CNN for plane localization and the dual CNN for cerebellum segmentation which are critical for accurate diagnostics during prenatal care. The differential CNN incorporates six different convolutional operators to capture diverse features for precise localization of specific planes within images. The dual CNN architecture integrates both the original image and complementary information such as feature maps, to enhance segmentation accuracy for the cerebellum. The models are trained on annotated datasets of ultrasound images, validated, and tested on separate datasets. The effectiveness of the proposed CNN architectures is determined by performing the specific tasks of plane localization and cerebellum segmentation, respectively. The proposed models achieved high performances of 98.6% and 0.956% from accuracy and dice coefficient (DSC) compared to existing approaches in medical image analysis. The findings of this study have significant implications for the prenatal prediction of neurodevelopmental disorders, offering a promising advancement in prenatal care and early diagnostics. The custom CNN architectures tailored to the specific tasks of plane localization and cerebellum segmentation highlight the importance of task-specific model design in medical imaging. While the study acknowledges certain limitations and challenges, the power of deep learning is analyzed for marking the benefit of healthcare and neurodevelopmental disorder prediction during pregnancy.

Adaptive spatial-channel feature fusion and self-calibrated convolution for early maize seedlings counting in UAV images

Accurate counting of crop plants is essential for agricultural science, particularly for yield forecasting, field management, and experimental studies. Traditional methods are labor-intensive and prone to errors. Unmanned Aerial Vehicle (UAV) technology offers a promising alternative; however, varying UAV altitudes can impact image quality, leading to blurred features and reduced accuracy in early maize seedling counts. To address these challenges, we developed RC-Dino, a deep learning methodology based on DINO, specifically designed to enhance the precision of seedling counts from UAV-acquired images. RC-Dino introduces two innovative components: a novel self-calibrating convolutional layer named RSCconv and an adaptive spatial feature fusion module called ASCFF. The RSCconv layer improves the representation of early maize seedlings compared to non-seedling elements within feature maps by calibrating spatial domain features. The ASCFF module enhances the discriminability of early maize seedlings by adaptively fusing feature maps extracted from different layers of the backbone network. Additionally, transfer learning was employed to integrate pre-trained weights with RSCconv, facilitating faster convergence and improved accuracy. The efficacy of our approach was validated using the Early Maize Seedlings Dataset (EMSD), comprising 1,233 annotated images of early maize seedlings, totaling 83,404 individual annotations. Testing on this dataset demonstrated that RC-Dino outperformed existing models, including DINO, Faster R-CNN, RetinaNet, YOLOX, and Deformable DETR. Specifically, RC-Dino achieved improvements of 16.29% in Average Precision (AP) and 8.19% in Recall compared to the DINO model. Our method also exhibited superior coefficient of determination (R²) values across different datasets for seedling counting. By integrating RSCconv and ASCFF into other detection frameworks such as Faster R-CNN, RetinaNet, and Deformable DETR, we observed enhanced detection and counting accuracy, further validating the effectiveness of our proposed method. These advancements make RC-Dino particularly suitable for accurate early maize seedling counting in the field. The source code for RSCconv and ASCFF is publicly available at https://github.com/collapser-AI/RC-Dino, promoting further research and practical applications.

Multidimensional computed measurement for highly accurate PCBA defect detection

Accurate defect detection in industrial automated optical inspection (AOI) is crucial as it directly affects product quality and production efficiency. Although numerous techniques have been developed for industrial defect detection, most of them rely on single-texture image data. This dependence limits the accuracy and robustness of the defect detection due to inadequate optical source information. To overcome the problem of low accuracy owing to the lack of 3D topographic information, a multidimensional information fusion (MIF) module is proposed that fuses texture image and depth image features. The MIF module includes tailored mechanisms to fully extract complementary semantic information from space and channel dimensions. A hierarchical fusion strategy further improves feature integration by enabling higher-layer feature fusion via lower-layer Transformer blocks and efficiently removing redundant features. Afterward, feature extraction is performed on the fused feature map, and output is obtained. To enhance the detection accuracy, the position information mask (PIM) module is introduced for post-processing. The PIM module uses surface mount devices (SMDs) position data from Gerber files to create a position information mask. The mask helps filter out defects that are often misidentified owing to missing design information. The results of the comparative experiments demonstrate that the average accuracy of our method on the printed circuit board assembly (PCBA) defect dataset is 99.93%, which is 5.64% higher than that of conventional YOLOV5. Furthermore, a comprehensive ablation study is conducted to elucidate the contribution of the proposed MIF and PIM modules. It demonstrated that the present model serves as a valuable reference for PCBA surface defect detection.

Identifying Spatial Distribution of Urban Vitality Using Self-Organizing Feature Map Neural Network

As a vital component of urban planning, urban vitality profoundly affects the sustainable development and well-being of cities. Existing evaluation methods struggle to effectively explain the spatial distribution between nonlinear indicators while simultaneously considering geographical location and spatial attributes. How do we propose a research framework to address this nonlinear spatial distribution? This question is crucial for the study of urban vitality. To bridge this research gap, this paper proposes an SOFM neural network utilizing multisource geospatial big data to explore the spatial distribution of urban vitality. Our results showed the following: (1) Urban vitality in the five dimensions of concentration, functional diversity, contact opportunity, accessibility, and distance from border vacuums decreased from the core area to the periphery, except for building diversity, which exhibited an opposite trend. (2) The urban vitality of Beijing’s central areas primarily showed a circled spatial structure and extended along the Beijing Central Axis and Chang’an Avenue. Additionally, a 15 km radius serves as a significant threshold, encompassing clusters 0, 1, and 2, which align with an important circle delineated by the Master Plan of Beijing (2016–2035). The findings of our research serve as valuable insights for enhancing urban vitality and urban planning.

PDNet by Partial Deep Convolution: A Better Lightweight Detector

Model lightweighting is significant in edge computing and mobile devices. Current studies on fast network design mainly focuses on model computation compression and speedup. Many models aim to compress models by dealing with redundant feature maps. However, most of these methods choose to preserve the feature maps with simple manipulations and do not effectively reduce redundant feature maps. This paper proposes a new convolution module, PDConv, which compresses redundant feature maps to reduce network complexity and increase network width to maintain accuracy. PDConv (Partial Deep Convolution) outperforms traditional methods in handling redundant feature maps, particularly in deep networks. Its FLOPs are comparable to deep separable convolution but with higher accuracy. This paper proposes PDBottleNeck and PDC2f (Partial Deep CSPDarknet53 to 2-Stage FPN) and build the lightweight network PDNet for experimental validation using the PASCAL VOC dataset. Compared to the popular HorNet, our method achieves an improvement of more than 25% in FLOPs and 1.8% in mAP50:95 accuracy. On the CoCo2017 dataset, our large PDNet achieves a 0.5% improvement in mAP75 and lower FLOPs than the latest RepVit.

IncARMAG: A convolutional neural network with multi-level autoregressive moving average graph convolutional processing framework for medical image classification

IncARMAG: A convolutional neural network with multi-level autoregressive moving average graph convolutional processing framework for medical image classification

A Sliding Window for Data Reuse in Deep Convolution Operations to Reduce Bandwidth Requirements and Resource Utilization

Convolutional Neural Networks (CNNs) have demonstrated high accuracy in applications such as object detection, classification, and image processing. However, convolutional layers account for the majority of computations within CNNs. Typically, these layers are executed on GPUs, resulting in higher-power consumption and hindering lightweight deployment. This paper presents a design that deploys convolutional layers on FPGAs with adjustable parameters. In this FPGA deployment, a 4 × 4 3D sliding window is used to traverse the data, reducing bandwidth requirements and facilitating seamless integration with subsequent processing stages. A three-dimensional plane buffer design is proposed, which implements data reuse. Compared to directly inputting the feature map and performing the computation, it reduces the on-chip memory bandwidth requirement by 75%. Additionally, a new addressing strategy is introduced to map 3D feature maps to RAM addresses, eliminating addressing time. Due to the resource-intensive nature of high-level synthesis (HLS) technology, HDL design is used for the convolutional layers. This design achieves an inference speed of 121.36 GOPS at a 16-bit width, providing a 39.10 times increase in performance compared to CPU implementations.

EGSDK-Net: Edge-Guided Stepwise Dual Kernel Update Network for Panoptic Segmentation

In recent years, panoptic segmentation has garnered increasing attention from researchers aiming to better understand scenes in images. Although many excellent studies have been proposed, they share some common unresolved issues. Firstly, panoptic segmentation, as a novel task, is still confined within inherent frameworks. Secondly, the prevalent kernel update strategies do not adequately utilize the information from each stage. To address these two issues, redwe propose an edge-guided stepwise dual kernel update network (EGSDK-Net) for panoptic segmentation; the core components are the real-time edge guidance module and the stepwise dual kernel update module. The first component, after extracting and positioning edge features through an extra branch, applies these features to the normally transmitted feature maps within the network to highlight the edges. The input image is initially processed with the Canny edge detector to generate and store the predicted edge map, which acts as the ground truth for supervising the extracted edge feature map. The stepwise dual kernel update module enhances the utilization of information by allowing each stage to update both its own kernel and that of the subsequent stage, thereby improving the judgment capabilities of the kernels. redEGSDK-Net achieves a PQ of 60.6, representing a 2.19% improvement over RT-K-Net.

Deep Feature Blend Attention: A New Frontier in Super Resolution Image Generation

Deep Feature Blend Attention: A New Frontier in Super Resolution Image Generation

Defects, microstructure, and properties in laser powder bed fusion IN718: Power density effects and feature maps

Defects, microstructure, and properties in laser powder bed fusion IN718: Power density effects and feature maps

FGA-YOLO: A one-stage and high-precision detector designed for fine-grained aircraft recognition

FGA-YOLO: A one-stage and high-precision detector designed for fine-grained aircraft recognition

State of Health Estimation of Lithium-Ion Batteries Using Data Augmentation and Feature Mapping

State of Health Estimation of Lithium-Ion Batteries Using Data Augmentation and Feature Mapping

Efficient and accurate feature-aided active tracking for underwater small targets in highly cluttered harbor environments using a full motion acoustic flow field solution.

To address the issue of tracking highly maneuverable small underwater intruders in the dynamic and heavily interfered environment of harbors, a local sparse motion acoustic flow (LSMAF) computation method constrained by robust high-order flux tensor (RHO-FT) feature has been proposed, along with a LSMAF feature-aided active tracking method. First, potential targets are localized from the dense dynamic clutter background of active sonar echographs using RHO-FT feature maps. Subsequently, on the basis of a dense motion acoustic flow calculation method, a local motion consistency criterion is introduced to the potential target area, establishing a method for calculating LSMAF, which updates the precise motion vectors of potential targets in real time. Finally, all measurements obtained from the RHO-FT feature map are iteratively filtered together with their corresponding motion vectors to maintain the kinematic consistency of the potential target's trajectory. Experiments conducted on a series of cooperative targets in real-world harbor environments show that LSMAF-aided tracking method outperforms the latest developments of tracking for underwater targets in terms of both tracking efficiency and trajectory accuracy.

A Robust Approach to Object Detection in Images using Improved T_CenterNet Method

Aims: A Robust Approach to Object Detection in Images using Improved T_CenterNet Method Background: Currently, two-stage image object detectors are the most common method of detecting objects. However, due to the slow processing of two-stage detectors, it is not always possible to apply them in real-time. Objective: To develop a more robust and effective method for detecting objects in images by building upon and improving the T_CenterNet algorithm Method: In this research paper, we propose CenterNet, a method based on a one-stage detector that takes into consideration the heat map branch of CenterNet, which is the most accurate anchorfree detector. To improve the outcomes of the upcoming image as well as detection efficiency, we disseminate the previously accurate long-term detection using a heatmap and a cascade guiding framework. Result: On the COCO dataset, our technique obtains detection outcome AP50 -65.5%, APS (Small Objects)-33.8, APM (Medium Objects)-47.0 at 38 frames per second. The results of the study show the superiority of the suggested approach over the conventional one. Conclusion: In this research, we have introduced an enhanced approach to image object detection using the improved T_CenterNet method. While the original CenterNet algorithm demonstrated superior precision compared to other single-stage algorithms, it suffered from slow detection speed. Our proposed approach, the Hourglass-208 model, built upon CenterNet, significantly improved detection performance. The proposed T_CenterNet algorithm improves upon the original CenterNet by incorporating the Hourglass backbone and feature map fusion techniques, enabling enhanced object detections. Additionally, we applied the Smooth L1 loss function to object size estimation within T_CenterNet, greatly enhancing overall detection performance by improving object size regression.

Complexities of feature-based learning systems, with application to reservoir computing

This paper studies complexity measures of reservoir systems. For this purpose, a more general model that we call a feature-based learning system, which is the composition of a feature map and of a final estimator, is studied. We study complexity measures such as growth function, VC-dimension, pseudo-dimension and Rademacher complexity. On the basis of the results, we discuss how the unadjustability of reservoirs and the linearity of readouts can affect complexity measures of the reservoir systems. Furthermore, some of the results generalize or improve the existing results.

Geometry-Aware Attenuation Learning for Sparse-View CBCT Reconstruction.

Cone Beam Computed Tomography (CBCT) plays a vital role in clinical imaging. Traditional methods typically require hundreds of 2D X-ray projections to reconstruct a high-quality 3D CBCT image, leading to considerable radiation exposure. This has led to a growing interest in sparse-view CBCT reconstruction to reduce radiation doses. While recent advances, including deep learning and neural rendering algorithms, have made strides in this area, these methods either produce unsatisfactory results or suffer from time inefficiency of individual optimization. In this paper, we introduce a novel geometry-aware encoder-decoder framework to solve this problem. Our framework starts by encoding multi-view 2D features from various 2D X-ray projections with a 2D CNN encoder. Leveraging the geometry of CBCT scanning, it then back-projects the multi-view 2D features into the 3D space to formulate a comprehensive volumetric feature map, followed by a 3D CNN decoder to recover 3D CBCT image. Importantly, our approach respects the geometric relationship between 3D CBCT image and its 2D X-ray projections during feature back projection stage, and enjoys the prior knowledge learned from the data population. This ensures its adaptability in dealing with extremely sparse view inputs without individual training, such as scenarios with only 5 or 10 X-ray projections. Extensive evaluations on two simulated datasets and one real-world dataset demonstrate exceptional reconstruction quality and time efficiency of our method.