Bottleneck Block Research Articles

Sparse-attention augmented domain adaptation for unsupervised person re-identification

The domain gap persists as a demanding problem for unsupervised domain adaptive person re-identification (UDA re-ID). In response to this question, we present a novel Sparse self-Attention Augmented Domain Adaptation approach (SAADA Model) to promote network performance. In this work, we put forward a composite computational primitive (SAAP). The SAAP leverages sparse self-attention and convolution to enhance domain adaptation at the primitive level. Using SAAP as a core component, we construct an augmented bottleneck block to improve domain adaptation at the bottleneck block level. Finally, the augmented bottleneck block for domain adaptation can be cascaded into the SAADA module. After extensive experiments for UDA re-ID benchmarks, we deploy the SAADA module one time after the stage corresponding to the minimum feature map, and the performance of this method exceeds some SOTA methods. For example, the mAP has increased by 5.1% from the Market-1501 to the difficult MSMT17.

CIS-UNet: Multi-class segmentation of the aorta in computed tomography angiography via context-aware shifted window self-attention

Advancements in medical imaging and endovascular grafting have facilitated minimally invasive treatments for aortic diseases. Accurate 3D segmentation of the aorta and its branches is crucial for interventions, as inaccurate segmentation can lead to erroneous surgical planning and endograft construction. Previous methods simplified aortic segmentation as a binary image segmentation problem, overlooking the necessity of distinguishing between individual aortic branches. In this paper, we introduce Context-Infused Swin-UNet (CIS-UNet), a deep learning model designed for multi-class segmentation of the aorta and thirteen aortic branches. Combining the strengths of Convolutional Neural Networks (CNNs) and Swin transformers, CIS-UNet adopts a hierarchical encoder–decoder structure comprising a CNN encoder, a symmetric decoder, skip connections, and a novel Context-aware Shifted Window Self-Attention (CSW-SA) module as the bottleneck block. Notably, CSW-SA introduces a unique adaptation of the patch merging layer, distinct from its traditional use in the Swin transformers. CSW-SA efficiently condenses the feature map, providing a global spatial context, and enhances performance when applied at the bottleneck layer, offering superior computational efficiency and segmentation accuracy compared to the Swin transformers. We evaluated our model on computed tomography (CT) scans from 59 patients through a 4-fold cross-validation. CIS-UNet outperformed the state-of-the-art Swin UNetR segmentation model by achieving a superior mean Dice coefficient of 0.732 compared to 0.717 and a mean surface distance of 2.40 mm compared to 2.75 mm. CIS-UNet’s superior 3D aortic segmentation offers improved accuracy and optimization for planning endovascular treatments. Our dataset and code will be made publicly available at https://github.com/mirthAI/CIS-UNet.

A lightweight parallel attention residual network for tile defect recognition

In modern industrial production, permanent magnet motors are an indispensable part of industrial manufacturing. The quality of the magnetic tiles directly affects the working performance of the permanent magnet motors, making the detection of defects on the surface of magnetic tiles critically important. However, due to the small size of defects on the tile image and the reflectivity of the defective surface, the details of image characteristics are not prominently acquired.These problems bring a lot of difficulties for the recognition of magnetic tile defects. In this paper, a magnetic tile defect detection method is proposed for the probAlems of unclear image features and small defects. First, the image is processed using linear variation to enhance the image detail features. Then, by introducing the inverted bottleneck block structure in MobileNetV2, the Attention Parallel Residual Convolution Block (APR) is proposed, and the Lightweight Parallel Attention Residual Network (LPAR-Net) is built. In APR Block, 7 × 7 convolution is introduced so that the model can extract spatial features from a larger range, and weighted fusion of input images by residual structure. In addition, in this paper, CBAM is improved, split into two parts and inserted into APR Block. Finally, the mainstream image classification models and the LPAR-Net proposed in this paper are used for comparison, respectively. The experimental results show that the method achieves 93.63% accuracy on the adopted dataset, which is better than the existing mainstream image classification network models DenseNet, MobileNet, ConvNext and so on. In addition, this paper introduces a strip steel surface defect dataset and compares it with the above image classification model, which verifies that the detection method proposed in this paper still has strong recognition capability.

YOLO-FMDI: A Lightweight YOLOv8 Focusing on a Multi-Scale Feature Diffusion Interaction Neck for Tomato Pest and Disease Detection

At the present stage, the field of detecting vegetable pests and diseases is in dire need of the integration of computer vision technologies. However, the deployment of efficient and lightweight object-detection models on edge devices in vegetable cultivation environments is a key issue. To address the limitations of current target-detection models, we propose a novel lightweight object-detection model based on YOLOv8n while maintaining high accuracy. In this paper, (1) we propose a new neck structure, Focus Multi-scale Feature Diffusion Interaction (FMDI), and inject it into the YOLOv8n architecture, which performs multi-scale fusion across hierarchical features and improves the accuracy of pest target detection. (2) We propose a new efficient Multi-core Focused Network (MFN) for extracting features of different scales and capturing local contextual information, which optimizes the processing power of feature information. (3) We incorporate the novel and efficient Universal Inverted Bottleneck (UIB) block to replace the original bottleneck block, which effectively simplifies the structure of the block and achieves the lightweight model. Finally, the performance of YOLO-FMDI is evaluated through a large number of ablation and comparison experiments. Notably, compared with the original YOLOv8n, our model reduces the parameters, GFLOPs, and model size by 18.2%, 6.1%, and 15.9%, respectively, improving the mean average precision (mAP50) by 1.2%. These findings emphasize the excellent performance of our proposed model for tomato pest and disease detection, which provides a lightweight and high-precision solution for vegetable cultivation applications.

Intelligent identification of rice leaf disease based on YOLO V5-EFFICIENT

Rice leaf diseases is a threat to sustainable rice production. Many methods via machine learning or artificial intelligence have been used to detect rice leaf diseases. However, these methods may fail to identify them or are slow in recognition. Therefore, a YOLO V5-IMPROVEMENT model is proposed to overcome these issues. Based on YOLO V5s, the improved K-Means is utilized to generate initial anchor sizes which more fit in the experimental dataset. An attention mechanism is added to let the feature extraction layer focus more on Regions of Interest (ROIs). This improvement enhances the efficiency of feature information extraction from objects of different scales. To better detect occluded targets, the loss function is improved to strengthen the regression effect of the predicted bounding boxes. Main branch gradient flow BottleNeck Block module of Cross Stage Partial (CSP) in the Neck is improved. The Spatial Pyramid Pooling-Fast (SPPF) is replaced with the S-SPPF module to optimize the model structure and ensure real-time performance. An ablation study was conducted and the proposed YOLO V5-EFFICIENT was compared with other models. Compared with YOLO V5, the detection result of YOLO V5-EFFICIENT increased by 9.90% in terms of mAP. Compared with other approaches, the improved model can better balance detection accuracy and speed when recognizing fine features of lesions. And it is capable of handling rice leaf disease detection tasks with numerous small targets. It is suggested that YOLO V5-EFFICIENT is of high accuracy and robustness.

Stem–Leaf segmentation and phenotypic trait extraction of individual plant using a precise and efficient point cloud segmentation network

Rapid and precise 3D organ segmentation is crucial for the automatic extraction of phenotypic traits, forming a fundamental prerequisite for intelligent plant breeding. The advancement of deep learning technology has replaced labor-intensive manual measurements and traditional computer vision methods, which are sensitive to parameters, in phenotypic trait extraction. However, current larger network structures not only require extensive point cloud data but also consume substantial computational resources, rendering them unsuitable for agricultural tasks with limited plant samples. Therefore, this study developed a lightweight 3D deep learning network (PEPNet) that achieves precise plant organ segmentation and stem-leaf phenotypic trait extraction. The adopted simple-but-effective network architecture and innovative modern operations, including a high-dimensional feature mapping strategy for preprocessing input points, a local feature extraction module based on inverted residual bottleneck block, and a cost-free attention block for spatial feature fusion, effectively implement multi-scale hierarchies and adaptively reduce computational overheads. Experimental results from cotton stem-leaf segmentation demonstrated that PEPNet not only presented approximately 2 × faster inference speed (9.59 ms) and throughput (146.32 plants per second) but also achieved competitive segmentation performance compared to other six state-of-the-art deep learning networks, namely PointNet++, DGCNN, CurveNet, Point Cloud Transformer, PointMLP, and SPoTr, achieving 95.99 %, 94.66 %, 95.32 %, and 91.31 % in Precision, Recall, F1-score, and mIoU, respectively. In transferability experiments with tomato and soybean plants, PEPNet achieved almost all the best metrics and significantly outperformed the second-best model (CurveNet). Furthermore, ablation study verified the optimal trade-off between efficiency and accuracy in this network. Any modifications to the modules could potentially disrupt the optimal trade-off. This work could contribute to reducing computational resources and annotation costs for applying segmentation methods in high-throughput phenotyping tasks.

WildARe-YOLO: A lightweight and efficient wild animal recognition model

For the protection of endangered species and successful wildlife population monitoring, wild animal recognition is essential. While deep learning models like YOLOv5 have shown promise in real-time object recognition, their practical applicability may be constrained by their high processing requirements. In this paper, we suggest a faster and lighter version of YOLOv5s for wild animal recognition. To lower computational costs for model parameters and floating-point operations (FLOPs) for the backbone, our suggested model includes Mobile Bottleneck Block modules and an improved StemBlock. We also use Focal-EIoU as a loss function to gauge the accuracy of the predicted bounding boxes during inference and employ a BiFPN-based neck. We tested our technique on three datasets, including Wild Animal Facing Extinction, Fishmarket, and MS COCO 2017. Additionally, our technique is compared with state-of-the-art deep learning models, and from the baseline model we recorded a 17.65% increase in FPS, 28.55% model parameters reduction, and 50.92% in FLOPs reduction. Furthermore, our model has a faster model loading time, which is critical for deployment in remote areas. This enables real-time species recognition on basic hardware, aiding conservation efforts through rapid analysis. The model advances deep learning in ecology by balancing efficiency with performance.

Interpretable Dual-branch EMGNet: A transfer learning-based network for inter-subject lower limb motion intention recognition

Currently, the fusion of surface Electromyography (EMG) and deep learning is gradually showing immense potential in the research of Lower Limb Motion Intention Recognition (LLMIR). Nevertheless, most deep learning algorithms have poor interpretability without special design or the help of other post-hoc analysis tools, as well as unsatisfactory performance in cross-subject prediction. Hence, this paper presents a novel Interpretable Dual-Branch EMG Network (IDB-EMGNet), in which one branch is dedicated to lower limb motion recognition, and the other is able to predict knee joint angles in advance. The shallow feature extraction module of IDB-EMGNet is constructed using an ante-hoc interpretable SincNet technique, which enables the detection of the spectral range of EMG used for the LLMIR task. An improved bottleneck block with shuffle attention is designed for deep feature extraction, which enhances model performance with only a little increase in complexity. The performance of IDB-EMGNet in both intra-subject and inter-subject scenarios is investigated, where the latter integrates the transfer learning technique. Specifically, by conducting model pre-training on source-domain subjects and transferring the learned knowledge to target-domain subjects, satisfactory performance can be achieved even with less computing resource. Experimental results on two publicly available datasets indicate that the proposed approach exhibits superior applicability to both normal subjects and knee-pathology patients, showing a promising prospect in the controller design of human-robot collaborative exoskeletons.

Cross channel weight sharing for image classification

Hypercomplex convolutional neural networks (HCNNs) have recently been used to improve deep learning architectures due to their ability to share weights across input channels and thus improve the cohesiveness of learned representations within the layers. Much work has been done to study the effect of hypercomplex representation in CNNs. However, to date, none of these models have used fully hypercomplex architectures, meaning that some of the layers in these earlier networks used conventional real-valued calculations that did not engage cross-channel weight sharing. Here, we introduce and study full hypercomplex CNNs by ensuring all layers perform hypercomplex calculations. For the earlier HCNNs, the real-valued layers are found in: (I) in the front end of the networks, (II) in the back end of the networks, and (III) in the residual blocks of the networks. The present research examines the performance of HCNNs when the modules mentioned above are replaced with hypercomplex equivalents. These representational networks have outperformed and shown state-of-the-art results compared to previous hypercomplex models (on some specific datasets). A disadvantage of these representational networks is that they consume high computational costs. To reduce this cost, novel separable hypercomplex networks (SHNNs) are proposed. They are created by factoring a quaternion convolutional module into two consecutive separable vectormap convolutional modules. As the successive layers of deep HCNNs perform local hierarchical grouping on increasingly abstract features, these groupings are responsible for the long-distance interaction problems found in the dense layers. To handle these problems, researchers have applied attention mechanisms in hypercomplex space.This paper offers a perspective on the basic concepts of representational networks and attention-based representational networks with the focus on (1) reviewing deep full HCNNs, including hypercomplex-based fully connected dense layer; (2) analyzing hypercomplex-based separable concept; (3) describing how the hypercomplex spatial layer of residual bottleneck block is replaced with an attention layer; and (4) providing a comprehensive summary comparing the performance of recent deep HCNNs, deep convolutional neural networks, and original attention based networks. We compare the performance of the networks in terms of accuracy and the number of trainable parameters on several image classification datasets.

A NEW INTELLIGENT MODEL BASED ON IMPROVED INCEPTION-V3 FOR ORAL CANCER AND CYST CLASSIFICATION

Oral cancer, which is also called mouth cancer, is cancer of the lining of the mouth, lips, or upper throat that has appeared in more than 355,000 people worldwide and caused more than 177,000 deaths, so it is essential to diagnose it as early as possible. Computed tomography (CT) scan is conducive to oral cancer diagnosis, but classifying oral CT images to cancer and cyst manually is difficult and time-consuming. A novel intelligent model based on improved Inception-v3 for classifying oral cancer and cyst CT images automatically is proposed in this paper. We replace the conventional convolution block in Inception-v3 with the Inverted Bottleneck Block and introduce Squeeze-and-Excitation Block (SEB) and Convolutional Block Attention Block (CBAB). The proposed model in this paper is trained on a dataset consisting of CT images of two classes (oral cancer and cyst), and the proposed model achieves 84.053% accuracy, 82.364% sensitivity, 84.508% specificity for oral cancer classification and outperforms other common models in classifying oral CT images.

LKDA-GAN: Cross-modality image synthesis via Generative Adversarial Network aggregating large kernel decomposable attention bottleneck block

Background and Objective:Mainstream image synthesis methods fail to capture local contextual information long-range dependence, and adaptability, especially under the influence of computation overload and random noise distribution, leading to the loss of contrast and detailed information. Approach:To alleviate these issues, we propose a Generative Adversarial Network aggregating large kernel decomposable attention (LKDA-GAN) bottleneck block for cross-modality image synthesis. Initially, a novel LKDA module is proposed by combining a spatial local convolution, a spatial long-range convolution, and a channel convolution, which aims to balance the local contextual information and the long-range dependence, and enhance the channel adaptability by enlarging the receptive field. Subsequently, a bottleneck block is designed and integrated into the LKDA module by feature dimensional transformation to capture rich semantic information and alleviate computational overhead. Ultimately, an auxiliary registration network (ARN) based on the noise transition matrix is put forward to learn the prior knowledge of the noise distribution and produce the unique optimal solution. Furthermore, it is based on Res-UNet to avoid network degradation. Main results:By analyzing qualitative assessment and quantitative measurement, extensive experiments demonstrate the proposed approach outperforms the state-of-the-art methods, and ablation studies also show the superiority of LKDA. Significance:LKDA-GAN provides a suitable way to synthesize images between different modalities, which is conductive to indicate disease areas and improve diagnosis accuracy.

SMTF: Sparse transformer with multiscale contextual fusion for medical image segmentation

Medical image segmentation aims at recognizing the object of interest from surrounding tissues and structures, which is essential for the reliable diagnosis and morphological analysis of specific lesions. Automatic medical image segmentation has been significantly boosted by deep Convolutional Neural Networks (CNNs). However, CNNs usually fail to model long-range interactions due to the intrinsic locality of convolutional operations, which limits the segmentation performance. Recently, Transformer has been successfully applied in various computer visions, which leverages the self-attention mechanism for modelling long-range interactions to capture global information. Nevertheless, self-attention suffers from lacks of spatial locality and efficient computation. To address these issues, in this work, we develop a new sparse medical Transformer (SMTF) with multiscale contextual fusion for medical image segmentation. The proposed model combines convolutional operations and attention mechanisms to form a U-shaped framework to capture both local and global information. Specifically, to reduce the computational cost of traditional Transformer, we design a novel sparse attention module to construct Transformer layers by spherical Locality Sensitive Hashing method. The sparse attention partitions the feature space into different attention buckets, and the attention calculation is conducted only in the individual bucket. The designed sparse Transformer layer further incorporates a bottleneck block to construct the encoder in SMTF. It is worth noting that the proposed sparse Transformer can also aggregate the global feature information in early stages, which enables the model to learn more local and global information by incorporating CNNs at lower layers. Furthermore, we introduce a deep supervision strategy to guide the model to fuse multiscale feature information. It further enables the SMTF to effectively propagate feature information across layers to preserve more input spatial information and mitigate information attenuation. Benefiting from these, it can achieve better segmentation performance while being more robust and efficient. The proposed SMTF is evaluated on multiple medical image segmentation datasets and a clinical nasopharyngeal carcinoma dataset. Extensive experiments have demonstrated its superiority on both qualitative and quantitative evaluations. Code and models are available at https://github.com/qmx717/sparse-attention.git.

TransBoNet: Learning camera localization with Transformer Bottleneck and Attention

6DoF camera localization is an important component of autonomous driving and navigation. Deep learning has achieved impressive results in localization, but its robustness in dynamic environments has not been adequately addressed. In this paper, we propose a framework based on hybrid attention mechanism which can be generally applied to existing CNN-based pose regressors to improve their robustness in dynamic environments. Specifically, we propose a novel Transformer Bottleneck (TBo) block including convolution, channel attention, and a position-aware self-attention mechanism, which extracts more geometrically robust features by capturing the corresponding long-term dependencies between pixels. Furthermore, we introduce shuffle attention (SA) before the pose regressor, which integrates feature information in both spatial and channel dimensions, forcing the network to learn geometrically robust features, reducing the effects of dynamic objects and illumination conditions to improve camera localization accuracy. We evaluate our method on commonly benchmarked indoor and outdoor datasets and the experimental results show that our proposed method can significantly improve localization performance compared compare favorably to contemporary pose regressors schemes. In addition, extensive ablation evaluations are conducted to prove the effectiveness of our proposed hybrid attention bottleneck block for pose regression networks.

YOLO-MBBi: PCB Surface Defect Detection Method Based on Enhanced YOLOv5

Printed circuit boards (PCBs) are extensively used to assemble electronic equipment. Currently, PCBs are an integral part of almost all electronic products. However, various surface defects can still occur during mass production. An enhanced YOLOv5s network named YOLO-MBBi is proposed to detect surface defects on PCBs to address the shortcomings of the existing PCB surface defect detection methods, such as their low accuracy and poor real-time performance. YOLO-MBBi uses MBConv (mobile inverted residual bottleneck block) modules, CBAM attention, BiFPN, and depth-wise convolutions to substitute layers in the YOLOv5s network and replace the CIoU loss function with the SIoU loss function during training. Two publicly available datasets were selected for this experiment. The experimental results showed that the mAP50 and recall values of YOLO-MBBi were 95.3% and 94.6%, which were 3.6% and 2.6% higher than those of YOLOv5s, respectively, and the FLOPs were 12.8, which was much smaller than YOLOv7’s 103.2. The FPS value reached 48.9. Additionally, after using another dataset, the YOLO-MBBi metrics also achieved satisfactory accuracy and met the needs of industrial production.

LMFFNet: A Well-Balanced Lightweight Network for Fast and Accurate Semantic Segmentation.

Real-time semantic segmentation is widely used in autonomous driving and robotics. Most previous networks achieved great accuracy based on a complicated model involving mass computing. The existing lightweight networks generally reduce the parameter sizes by sacrificing the segmentation accuracy. It is critical to balance the parameters and accuracy for real-time semantic segmentation. In this article, we propose a lightweight multiscale-feature-fusion network (LMFFNet) mainly composed of three types of components: split-extract-merge bottleneck (SEM-B) block, feature fusion module (FFM), and multiscale attention decoder (MAD), where the SEM-B block extracts sufficient features with fewer parameters. FFMs fuse multiscale semantic features to effectively improve the segmentation accuracy and the MAD well recovers the details of the input images through the attention mechanism. Without pretraining, LMFFNet-3-8 achieves 75.1% mean intersection over union (mIoU) with 1.4 M parameters at 118.9 frames/s using RTX 3090 GPU. More experiments are investigated extensively on various resolutions on other three datasets of CamVid, KITTI, and WildDash2. The experiments verify that the proposed LMFFNet model makes a decent tradeoff between segmentation accuracy and inference speed for real-time tasks. The source code is publicly available at https://github.com/Greak-1124/LMFFNet.

Gaze Estimation via Strip Pooling and Multi-Criss-Cross Attention Networks

Deep learning techniques for gaze estimation usually determine gaze direction directly from images of the face. These algorithms achieve good performance because face images contain more feature information than eye images. However, these image classes contain a substantial amount of redundant information that may interfere with gaze prediction and may represent a bottleneck for performance improvement. To address these issues, we model long-distance dependencies between the eyes via Strip Pooling and Multi-Criss-Cross Attention Networks (SPMCCA-Net), which consist of two newly designed network modules. One module is represented by a feature enhancement bottleneck block based on fringe pooling. By incorporating strip pooling, this residual module not only enlarges its receptive fields to capture long-distance dependence between the eyes but also increases weights on important features and reduces the interference of redundant information unrelated to gaze. The other module is a multi-criss-cross attention network. This module exploits a cross-attention mechanism to further enhance long-range dependence between the eyes by incorporating the distribution of eye-gaze features and providing more gaze cues for improving estimation accuracy. Network training relies on the multi-loss function, combined with smooth L1 loss and cross entropy loss. This approach speeds up training convergence while increasing gaze estimation precision. Extensive experiments demonstrate that SPMCCA-Net outperforms several state-of-the-art methods, achieving mean angular error values of 10.13° on the Gaze360 dataset and 6.61° on the RT-gene dataset.

Multi-Scale Region-Of-Interest based Deep Learning for fruit disease identification and classification

Agricultural productivity is mainly affected by the different types of fruit diseases. To reduce such impacts, automated fruit disease identification and categorization model is vital. So, a Multi-scale Deep Learning (MDL) model was designed to identify and categorize the variety of strawberry fruit diseases. This model has an Adaptive Receptive Field Module (ARFM), a Bottleneck Block (BB), Depth-wise Residual Blocks (DRBs) and the Squeeze Excitation Residual Network (SE-ResNet) units to extract the Feature Maps (FMs) related to all disease classes. But, this model cannot partition individual fruit samples. Also, the number of fruits with multiple infections within a particular image was not easily captured. So, this article develops a Multi-scale Region-Of-Interest (ROI)-based DL model called (MRDL) to improve the accuracy of identification and categorization of a single fruit sample with a certain disease in a real-time scenario. This model is performed in 2 different phases. First, more robust features are extracted by the DRB, ARFM, and BB in the initial phase. Then, an ROI Fused (ROIF) unit, the SE-ResNet unit, a classification unit, and a mask unit are involved in the second phase. The ROIF unit obtains the FMs for all candidate ROIs using the candidate bounding boxes from the initial phase. The extracted FMs per sampled position are aggregated to get the fused FM, which is resized by integrating the attention strategy with the SE-ResNet unit. Those resized fused FMs are independently passed to the classification and mask units to classify the candidate bounding boxes and localize the infected pixels in the candidate ROIs, respectively. Finally, the test results exhibit that the MRDL model achieves 91.1%, 90.8%, and 91.6% accuracy on apple, citrus, and tomato disease databases, respectively compared to the classical models.

DCA-Net: Dual-branch contextual-aware network for auxiliary localization and segmentation of parathyroid glands

Hypoparathyroidism is one of the major complications after thyroid cancer surgery, which seriously affects patients’ quality of life. How to effectively protect parathyroid function and reduce the incidence of hypoparathyroidism has been a key research area in thyroid surgery. Before the operation, the identification and localization of the parathyroid glands by ultrasound images can help surgeons effectively protect parathyroid glands during thyroidectomy. Computer-aided diagnosis (CAD) systems are often considered useful tools to aid in medical image analysis examinations, which provide new solutions for preoperative identification and localization of the parathyroid glands in ultrasound images with high accuracy. To reduce the incidence of hypoparathyroidism and capture both local and global feature dependencies, we propose a dual-branch contextual-aware network (DCA-Net) with Transformer. Firstly, the Transformer as the long-range feature extractor branch is used as the input image patches as the input sequence for extracting the global context. Meanwhile, we build the residual bottleneck block and Transformer bottleneck on the feature encoder branch to locally aggregate the sampled features on different semantic scales. Finally, we use the channel and spatial fusion module (CSFM) to aggregate information from these two branches. The proposed DCA-Net can effectively mitigate the loss of details due to information recession and establish global and local feature dependencies by utilizing these complementing components. Experiments using our dataset of Ultrasound images of Parathyroid Glands reveal that our method outperforms state-of-the-art baselines consistently.

Assessing Environmental Oil Spill Based on Fluorescence Images of Water Samples and Deep Learning

Measuring oil concentration in the aquatic environment is essential for determining the potential exposure, risk, or injury for oil spill response and natural resource damage assessment. Conventional analytical chemistry methods require samples to be collected in the field, shipped, and processed in the laboratory, which is also rather time-consuming, laborious, and costly. For rapid field response immediately after a spill, there is a need to estimate oil concentration in near real time. To make the oil analysis more portable, fast, and cost effective, we developed a plug-and-play device and a deep learning model to assess oil levels in water using fluorescent images of water samples. We constructed a 3D-printed device to collect fluorescent images of solvent-extracted water samples using an iPhone. We prepared approximately 1,300 samples of oil at different concentrations to train and test the deep learning model. The model comprises a convolutional neural network and a novel module of histogram bottleneck block with an attention mechanism to exploit the spectral features found in low-contrast images. This model predicts the oil concentration in weight per volume based on fluorescence image. We devised a confidence interval estimator by combining gradient boosting and polymodal regressor to provide a confidence assessment of our results. Our model achieved sufficient accuracy to predict oil levels for most environmental applications. We plan to improve the device and iPhone application as a near-real-time tool for oil spill responders to measure oil in water.

A Fast and Accurate Human Pose Estimation Method Based on Multi-Scale Feature Fusion Grid Structure

Human pose estimation (HPE) is a research hotspot in the field of computer vision. Most of the existing approaches first generate low-resolution representation from high-resolution representation through continuous serial downsampling, and then reconstruct high-resolution results from low-resolution features through continuous serial upsampling, which loses a lot of effective feature information and leads to slow model inference. In this paper, the Fast Accuracy Network (FANet), a framework that enables fast and high-accuracy HPE, is proposed. The innovation lies in that, first of all, a grid structure is proposed and adopted, which can be regarded as a set of deep paths and shallow paths. The structure uses multiple high-resolution and low-resolution branch pairs to perform skip-level connections at different scale-space levels so that the information can be exchanged between different resolution representations for many times. The feature information fusion of multi-scale space is realized to obtain more abundant feature information. Second, an improved bottleneck block is proposed to extract effective feature information with fewer parameters, ensuring that the computational burden is reduced without sacrificing accuracy performance. The experimental results show that, compared with other current models, FANet has faster inference speed on the premise of a slight improvement in accuracy performance.