ABSTRACT
Regarding the difficulty of extracting the acquired fault signal features of bearings from a strong background noise vibration signal, coupled with the fact that one-dimensional (1D) signals provide limited fault information, an optimal time frequency fusion symmetric dot pattern (SDP) bearing fault feature enhancement and diagnosis method is proposed. Firstly, the vibration signals are transformed into two-dimensional (2D) features by the time frequency fusion algorithm SDP, which can multi-scale analyze the fluctuations of signals at minor scales, as well as enhance bearing fault features. Secondly, the bat algorithm is employed to optimize the SDP parameters adaptively. It can effectively improve the distinctions between various types of faults. Finally, the fault diagnosis model can be constructed by a deep convolutional neural network (DCNN). To validate the effectiveness of the proposed method, Case Western Reserve University's (CWRU) bearing fault dataset and bearing fault dataset laboratory experimental platform were used. The experimental results illustrate that the fault diagnosis accuracy of the proposed method is 100%, which proves the feasibility and effectiveness of the proposed method. By comparing with other 2D transformer methods, the experimental results illustrate that the proposed method achieves the highest accuracy in bearing fault diagnosis. It validated the superiority of the proposed methodology.
ABSTRACT
Pest infestation poses significant threats to grain storage due to pests' behaviors of feeding, respiration, excretion, and reproduction. Efficient pest detection and control are essential to mitigate these risks. However, accurate detection of small grain pests remains challenging due to their small size, high variability, low contrast, and cluttered background. Salient pest detection focuses on the visual features that stand out, improving the accuracy of pest identification in complex environments. Drawing inspiration from the rapid pest recognition abilities of humans and birds, we propose a novel Cascaded Aggregation Convolution Network (CACNet) for pest detection and control in stored grain. Our approach aims to improve detection accuracy by employing a reverse cascade feature aggregation network that imitates the visual attention mechanism in humans when observing and focusing on objects of interest. The CACNet uses VGG16 as the backbone network and incorporates two key operations, namely feature enhancement and feature aggregation. These operations merge the high-level semantic information and low-level positional information of salient objects, enabling accurate segmentation of small-scale grain pests. We have curated the GrainPest dataset, comprising 500 images showcasing zero to five or more pests in grains. Leveraging this dataset and the MSRA-B dataset, we validated our method's efficacy, achieving a structure S-measure of 91.9%, and 90.9%, and a weighted F-measure of 76.4%, and 91.0%, respectively. Our approach significantly surpasses the traditional saliency detection methods and other state-of-the-art salient object detection models based on deep learning. This technology shows great potential for pest detection and assessing the severity of pest infestation based on pest density in grain storage facilities. It also holds promise for the prevention and control of pests in agriculture and forestry.
ABSTRACT
Colorectal polyps serve as potential precursors of colorectal cancer and automating polyp segmentation aids physicians in accurately identifying potential polyp regions, thereby reducing misdiagnoses and missed diagnoses. However, existing models often fall short in accurately segmenting polyps due to the high degree of similarity between polyp regions and surrounding tissue in terms of color, texture, and shape. To address this challenge, this study proposes a novel three-stage polyp segmentation network, named Reverse Attention Feature Purification with Pyramid Vision Transformer (RAFPNet), which adopts an iterative feedback UNet architecture to refine polyp saliency maps for precise segmentation. Initially, a Multi-Scale Feature Aggregation (MSFA) module is introduced to generate preliminary polyp saliency maps. Subsequently, a Reverse Attention Feature Purification (RAFP) module is devised to effectively suppress low-level surrounding tissue features while enhancing high-level semantic polyp information based on the preliminary saliency maps. Finally, the UNet architecture is leveraged to further refine the feature maps in a coarse-to-fine approach. Extensive experiments conducted on five widely used polyp segmentation datasets and three video polyp segmentation datasets demonstrate the superior performance of RAFPNet over state-of-the-art models across multiple evaluation metrics.
Subject(s)
Colonic Polyps , Humans , Colonic Polyps/diagnostic imaging , Image Interpretation, Computer-Assisted/methods , AlgorithmsABSTRACT
Objective.We aim to develop a Multi-modal Fusion and Feature Enhancement U-Net (MFFE U-Net) coupling with stem cell niche proximity estimation to improve voxel-wise Glioblastoma (GBM) recurrence prediction.Approach.57 patients with pre- and post-surgery magnetic resonance (MR) scans were retrospectively solicited from 4 databases. Post-surgery MR scans included two months before the clinical diagnosis of recurrence and the day of the radiologicaly confirmed recurrence. The recurrences were manually annotated on the T1ce. The high-risk recurrence region was first determined. Then, a sparse multi-modal feature fusion U-Net was developed. The 50 patients from 3 databases were divided into 70% training, 10% validation, and 20% testing. 7 patients from the 4th institution were used as external testing with transfer learning. Model performance was evaluated by recall, precision, F1-score, and Hausdorff Distance at the 95% percentile (HD95). The proposed MFFE U-Net was compared to the support vector machine (SVM) model and two state-of-the-art neural networks. An ablation study was performed.Main results.The MFFE U-Net achieved a precision of 0.79 ± 0.08, a recall of 0.85 ± 0.11, and an F1-score of 0.82 ± 0.09. Statistically significant improvement was observed when comparing MFFE U-Net with proximity estimation couple SVM (SVMPE), mU-Net, and Deeplabv3. The HD95 was 2.75 ± 0.44 mm and 3.91 ± 0.83 mm for the 10 patients used in the model construction and 7 patients used for external testing, respectively. The ablation test showed that all five MR sequences contributed to the performance of the final model, with T1ce contributing the most. Convergence analysis, time efficiency analysis, and visualization of the intermediate results further discovered the characteristics of the proposed method.Significance. We present an advanced MFFE learning framework, MFFE U-Net, for effective voxel-wise GBM recurrence prediction. MFFE U-Net performs significantly better than the state-of-the-art networks and can potentially guide early RT intervention of the disease recurrence.
Subject(s)
Brain Neoplasms , Glioblastoma , Magnetic Resonance Imaging , Neoplasm Recurrence, Local , Humans , Glioblastoma/diagnostic imaging , Glioblastoma/pathology , Brain Neoplasms/diagnostic imaging , Brain Neoplasms/pathology , Neoplasm Recurrence, Local/diagnostic imaging , Image Processing, Computer-Assisted/methods , Neural Networks, Computer , Retrospective Studies , Recurrence , Male , Female , Middle AgedABSTRACT
BACKGROUND: Automatic segmentation of vertebrae in spinal x-ray images is crucial for clinical diagnosis, case analysis, and surgical planning of spinal lesions. PURPOSE: However, due to the inherent characteristics of x-ray images, including low contrast, high noise, and uneven grey scale, it remains a critical and challenging problem in computer-aided spine image analysis and disease diagnosis applications. METHODS: In this paper, a Multiscale Feature Enhancement Network (MFENet), is proposed for segmenting whole spinal x-ray images, to aid doctors in diagnosing spinal-related diseases. To enhance feature extraction, the network incorporates a Dual-branch Feature Extraction Module (DFEM) and a Semantic Aggregation Module (SAM). The DFEM has a parallel dual-branch structure. The upper branch utilizes multiscale convolutional kernels to extract features from images. Employing convolutional kernels of different sizes helps capture details and structural information at different scales. The lower branch incorporates attention mechanisms to further optimize feature representation. By modeling the feature maps spatially and across channels, the network becomes more focused on key feature regions and suppresses task-irrelevant information. The SAM leverages contextual semantic information to compensate for details lost during pooling and convolution operations. It integrates high-level feature information from different scales to reduce segmentation result discontinuity. In addition, a hybrid loss function is employed to enhance the network's feature extraction capability. RESULTS: In this study, we conducted a multitude of experiments utilizing dataset provided by the Spine Surgery Department of Henan Provincial People's Hospital. The experimental results indicate that our proposed MFENet demonstrates superior segmentation performance in spinal segmentation on x-ray images compared to other advanced methods, achieving 92.61 ± 0.431 for MIoU, 92.42 ± 0.329 for DSC, and 99.51 ± 0.037 for Global_accuracy. CONCLUSIONS: Our model is able to more effectively learn and extract global contextual semantic information, significantly improving spinal segmentation performance, further aiding doctors in analyzing patient conditions.
ABSTRACT
In view of the frequent failures occurring in rolling bearings, the strong background noise present in signals, weak features, and difficulties associated with extracting fault characteristics, a method of enhancing and diagnosing rolling bearing faults based on coarse-grained lattice features (CGLFs) is proposed. First, the vibrational signals of bearings are subjected to adaptive filtering to eliminate background noise. Second, frequency-domain transformation is performed, and a coarse-grained approach is used to continuously segment the spectrum. Within each segment, amplitude-enhancement operations are executed, transforming the data into a CGLF graph that enhances fault characteristics. This graph is then fed into a Swin Transformer-based pattern-recognition network. Third and finally, a high-precision fault diagnosis model is constructed using fully connected layers and Softmax, enabling the diagnosis of bearing faults. The fault recognition accuracy reaches 98.30% and 98.50% with public datasets and laboratory data, respectively, thereby validating the feasibility and effectiveness of the proposed method. This research offers an efficient and feasible fault diagnosis approach for rolling bearings.
ABSTRACT
Accurate risk prediction for myocardial infarction (MI) is crucial for preventive strategies, given its significant impact on global mortality and morbidity. Here, we propose a novel deep-learning approach to enhance the prediction of incident MI cases by incorporating metabolomics alongside clinical risk factors. We utilized data from the KORA cohort, including the baseline S4 and follow-up F4 studies, consisting of 1454 participants without prior history of MI. The dataset comprised 19 clinical variables and 363 metabolites. Due to the imbalanced nature of the dataset (78 observed MI cases and 1376 non-MI individuals), we employed a generative adversarial network (GAN) model to generate new incident cases, augmenting the dataset and improving feature representation. To predict MI, we further utilized multi-layer perceptron (MLP) models in conjunction with the synthetic minority oversampling technique (SMOTE) and edited nearest neighbor (ENN) methods to address overfitting and underfitting issues, particularly when dealing with imbalanced datasets. To enhance prediction accuracy, we propose a novel GAN for feature-enhanced (GFE) loss function. The GFE loss function resulted in an approximate 2% improvement in prediction accuracy, yielding a final accuracy of 70%. Furthermore, we evaluated the contribution of each clinical variable and metabolite to the predictive model and identified the 10 most significant variables, including glucose tolerance, sex, and physical activity. This is the first study to construct a deep-learning approach for producing 7-year MI predictions using the newly proposed loss function. Our findings demonstrate the promising potential of our technique in identifying novel biomarkers for MI prediction.
ABSTRACT
Introduction: Accurately counting the number of dense objects in an image, such as pedestrians or vehicles, is a challenging and practical task. The existing density map regression methods based on CNN are mainly used to count a class of dense objects in a single scene. However, in complex traffic scenes, objects such as vehicles and pedestrians usually exist at the same time, and multiple classes of dense objects need to be counted simultaneously. Methods: To solve the above issues, we propose a new multiple types of dense object counting method based on feature enhancement, which can enhance the features of dense counting objects in complex traffic scenes to realize the classification and regression counting of dense vehicles and people. The counting model consists of the regression subnet and the classification subnet. The regression subnet is primarily used to generate two-channel predicted density maps, mainly including the initial feature layer and the feature enhancement layer, in which the feature enhancement layer can enhance the classification features and regression counting features of dense objects in complex traffic scenes. The classification subnet mainly supervises classifying dense vehicles and people into two feature channels to assist the regression counting task of the regression subnets. Results: Our method is compared on VisDrone+ datasets, ApolloScape+ datasets, and UAVDT+ datasets. The experimental results show that the method counts two kinds of dense objects simultaneously and outputs a high-quality two-channel predicted density map. The counting performance is better than the state-of-the-art counting network in dense people and vehicle counting. Discussion: In future work, we will further improve the feature extraction ability of the model in complex traffic scenes to classify and count a variety of dense objects such as cars, pedestrians, and non-motor vehicles.
ABSTRACT
Sea ice, as an important component of the Earth's ecosystem, has a profound impact on global climate and human activities due to its thickness. Therefore, the inversion of sea ice thickness has important research significance. Due to environmental and equipment-related limitations, the number of samples available for remote sensing inversion is currently insufficient. At high spatial resolutions, remote sensing data contain limited information and noise interference, which seriously affect the accuracy of sea ice thickness inversion. In response to the above issues, we conducted experiments using ice draft data from the Beaufort Sea and designed an improved GBDT method that integrates feature-enhancement and active-learning strategies (IFEAL-GBDT). In this method, the incident angle and time series are used to perform spatiotemporal correction of the data, reducing both temporal and spatial impacts. Meanwhile, based on the original polarization information, effective multi-attribute features are generated to expand the information content and improve the separability of sea ice with different thicknesses. Taking into account the growth cycle and age of sea ice, attributes were added for month and seawater temperature. In addition, we studied an active learning strategy based on the maximum standard deviation to select more informative and representative samples and improve the model's generalization ability. The improved GBDT model was used for training and prediction, offering advantages in dealing with nonlinear, high-dimensional data, and data noise problems, further expanding the effectiveness of feature-enhancement and active-learning strategies. Compared with other methods, the method proposed in this paper achieves the best inversion accuracy, with an average absolute error of 8 cm and a root mean square error of 13.7 cm for IFEAL-GBDT and a correlation coefficient of 0.912. This research proves the effectiveness of our method, which is suitable for the high-precision inversion of sea ice thickness determined using Sentinel-1 data.
ABSTRACT
The transportation of hazardous chemicals on roadways has raised significant safety concerns. Incidents involving these substances often lead to severe and devastating consequences. Consequently, there is a pressing need for real-time detection systems tailored for hazardous material vehicles. However, existing detection methods face challenges in accurately identifying smaller targets and achieving high precision. This paper introduces a novel solution, HMV-YOLO, an enhancement of the YOLOv7-tiny model designed to address these challenges. Within this model, two innovative modules, CBSG and G-ELAN, are introduced. The CBSG module's mathematical model incorporates components such as Convolution (Conv2d), Batch Normalization (BN), SiLU activation, and Global Response Normalization (GRN) to mitigate feature collapse issues and enhance neuron activity. The G-ELAN module, building upon CBSG, further advances feature fusion. Experimental results showcase the superior performance of the enhanced model compared to the original one across various evaluation metrics. This advancement shows great promise for practical applications, particularly in the context of real-time monitoring systems for hazardous material vehicles.
ABSTRACT
The recognition of human activity is crucial as the Internet of Things (IoT) progresses toward future smart homes. Wi-Fi-based motion-recognition stands out due to its non-contact nature and widespread applicability. However, the channel state information (CSI) related to human movement in indoor environments changes with the direction of movement, which poses challenges for existing Wi-Fi movement-recognition methods. These challenges include limited directions of movement that can be detected, short detection distances, and inaccurate feature extraction, all of which significantly constrain the wide-scale application of Wi-Fi action-recognition. To address this issue, we propose a direction-independent CSI fusion and sharing model named CSI-F, one which combines Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU). Specifically, we have introduced a series of signal-processing techniques that utilize antenna diversity to eliminate random phase shifts, thereby removing noise influences unrelated to motion information. Later, by amplifying the Doppler frequency shift effect through cyclic actions and generating a spectrogram, we further enhance the impact of actions on CSI. To demonstrate the effectiveness of this method, we conducted experiments on datasets collected in natural environments. We confirmed that the superposition of periodic actions on CSI can improve the accuracy of the process. CSI-F can achieve higher recognition accuracy compared with other methods and a monitoring coverage of up to 6 m.
Subject(s)
Internet of Things , Movement , Humans , Motion , Doppler Effect , EnvironmentABSTRACT
Underwater light attenuation leads to decreased image contrast. This reduction in contrast subsequently decreases target visibility. Additionally, marine target detection is challenging due to multi-scale problems from varying target-to-device distances, complex target clustering, and noise from waterborne particulates.To address these issues, we propose MTD-YOLOv5.Initially, we enhance image contrast with grayscale equalization and mitigate color shift issues through color space transformation.We then introduce a novel feature extraction module, PCBR, combining max pooling and convolution layers for more effective target feature extraction from the background.Furthermore, we present the Multi-Scale Perceptual Hybrid Pooling (MHP) module.This module integrates horizontal and vertical receptive fields to establish long-range dependencies, thereby capturing hidden target information in deep network feature maps. In the Labeled Fishes in the Wild test datasets, MTD-YOLOv5 achieves a precision of 88.1% and a mean Average Precision (mAP[0.5:.95]) of 49.6%.These results represent improvements of 2.6% in precision and 0.4% in mAP over the original YOLOv5.
ABSTRACT
BACKGROUND: Segmenting colorectal polyps presents a significant challenge due to the diverse variations in their size, shape, texture, and intricate backgrounds. Particularly demanding are the so-called "camouflaged" polyps, which are partially concealed by surrounding tissues or fluids, adding complexity to their detection. METHODS: We present CPSNet, an innovative model designed for camouflaged polyp segmentation. CPSNet incorporates three key modules: the Deep Multi-Scale-Feature Fusion Module, the Camouflaged Object Detection Module, and the Multi-Scale Feature Enhancement Module. These modules work collaboratively to improve the segmentation process, enhancing both robustness and accuracy. RESULTS: Our experiments confirm the effectiveness of CPSNet. When compared to state-of-the-art methods in colon polyp segmentation, CPSNet consistently outperforms the competition. Particularly noteworthy is its performance on the ETIS-LaribPolypDB dataset, where CPSNet achieved a remarkable 2.3% increase in the Dice coefficient compared to the Polyp-PVT model. CONCLUSION: In summary, CPSNet marks a significant advancement in the field of colorectal polyp segmentation. Its innovative approach, encompassing multi-scale feature fusion, camouflaged object detection, and feature enhancement, holds considerable promise for clinical applications.
Subject(s)
Colonic Polyps , Humans , Colonic Polyps/diagnostic imaging , Colon , Image Processing, Computer-AssistedABSTRACT
Diagnosis of incipient faults of metro train bearings is a difficult problem under the double masking of strong wheel-rail impact interference and background noise. A novel feature extraction method using improved complementary complete local mean decomposition with adaptive noise (ICCELMDAN) and mixture correntropy-based adaptive feature enhancement (AFE) is proposed in this paper. The ICCELMDAN method uses a proposed complementary adaptive noise-assisted iterative sifting method to improve its anti-mixing and anti-splitting performance, and then can extract the complete feature from faulty bearing signals under strong background noise. The AFE method adaptively obtains the optimal parameters of mixture correntropy (MC) by employing a newly developed fault energy of mixture correntropy as the objective function in the marine predators algorithm (MPA), and can enhance the weak fault characteristic signal under strong wheel-rail impact interferences. The proposed method effectively combines the complete feature extraction capability of ICCELMDAN and the powerful feature enhancement capability of AFE, which can accurately diagnose the weak faults of metro train bearings under strong wheel-rail impact interferences in simulated and practical scenarios. Furthermore, it outperforms the existing methods in completeness of feature extraction, diagnosis accuracy and robustness from the comparative studies.
ABSTRACT
Blind deconvolution can remove the effects of complex paths and extraneous disturbances, thus recovering simple features of the original fault source, and is used extensively in the field of fault diagnosis. However, it can only identify and extract the statistical mean of the fault impact features in a single domain and is unable to simultaneously highlight the local features of the signal in the time-frequency domain. Therefore, the extraction effect of weak fault signals is generally not ideal. In this paper, a new time-frequency slice extraction method is proposed. The method first computes a high temporal resolution spectrum of the signal by short-time Fourier transform to obtain multiple frequency slices with distinct temporal waveforms. Subsequently, the constructed harmonic spectral feature index is used to quantify and target the intensity of feature information in each frequency slice and enhance their fault characteristics using maximum correlation kurtosis deconvolution. Enhancing the local features of selected frequency slice clusters can reduce noise interference and obtain signal components with more obvious fault signatures. Finally, the validity of the method was confirmed by a simulated signal and fault diagnosis of the rolling bearing outer and inner rings was accomplished sequentially. Compared with other common deconvolution methods, the proposed method obtains more accurate and effective results in identifying fault messages.
ABSTRACT
A proposed method, Enhancement, integration, and Expansion, aims to activate the representation of detailed features for occluded person re-identification. Region and context are two important and complementary features, and integrating them in an occluded environment can effectively improve the robustness of the model. Firstly, a self-enhancement module is designed. Based on the constructed multi-stream architecture, rich and meaningful feature interference is introduced in the feature extraction stage to enhance the model's ability to perceive noise. Next, a collaborative integration module similar to cascading cross-attention is proposed. By studying the intrinsic interaction patterns of regional and contextual features, it adaptively fuses features across streams and enhances the diverse and complete representation of internal information. The module is not only robust to complex occlusions, but also mitigates the feature interference problem due to similar appearances or scenes. Finally, a matching expansion module that enhances feature discriminability and completeness is proposed. Providing more stable and accurate features for recognition. Compared with state-of-the-art methods on two occluded and holistic datasets, the proposed method is proved to be advanced and the effectiveness of the module is proved by extensive ablation studies.
Subject(s)
Biometric Identification , Neural Networks, Computer , HumansABSTRACT
Deep learning methods exhibited significant advantages in mapping highly nonlinear relationships with acceptable computational speed, and have been widely used to predict water quality. However, various model selection and construction methods resulted in differences in prediction accuracy and performance. Hence, a unified deep learning framework for water quality prediction was established in the paper, including data processing module, feature enhancement module, and data prediction module. In the established model, the data processing module based on wavelet transform method was applied to decomposing complex nonlinear meteorology, hydrology, and water quality data into multiple frequency domain signals for extracting self characteristics of data cyclic and fluctuations. The feature enhancement module based on Informer Encoder was used to enhance feature encoding of time series data in different frequency domains to discover global time dependent features of variables. Finally, the data prediction module based on the stacked bidirectional long and short term memory network (SBiLSTM) method was employed to strengthen the local correlation of feature sequences and predict the water quality. The established model framework was applied in Lijiang River in Guilin, China. The maximum relative errors between the predicted and observed values for dissolved oxygen (DO), chemical oxygen demand (CODMn) were 12.4% and 20.7%, suggesting a satisfactory prediction performance of the established model. The validation results showed that the established model was superior to all other models in terms of prediction accuracy with RMSE values 0.329, 0.121, MAE values 0.217, 0.057, SMAPE values 0.022, 0.063 for DO and CODMn, respectively. Ablation tests confirmed the necessity and rationality of each module for the established model framework. The established method provided a unified deep learning framework for water quality prediction.
Subject(s)
Deep Learning , Water Quality , China , Hydrology , Meteorology , OxygenABSTRACT
The current semantic segmentation algorithms suffer from encoding feature distortion and small object feature loss. Context information exchange can effectively address the feature distortion problem, but it has the issue of fixed spatial range. Maintaining the input feature resolution can reduce the loss of small object information but would slow down the network's operation speed. To tackle these problems, we propose a lightweight semantic segmentation network with configurable context and small object attention (CCSONet). CCSONet includes a long-short distance configurable context feature enhancement module (LSCFEM) and a small object attention decoding module (SOADM). The LSCFEM differs from the regular context exchange module by configuring long and short-range relevant features for the current feature, providing a broader and more flexible spatial range. The SOADM enhances the features of small objects by establishing correlations among objects of the same category, avoiding the introduction of redundancy issues caused by high-resolution features. On the Cityscapes and Camvid datasets, our network achieves the accuracy of 76.9 mIoU and 73.1 mIoU, respectively, while maintaining speeds of 87 FPS and 138 FPS. It outperforms other lightweight semantic segmentation algorithms in terms of accuracy.
ABSTRACT
Aiming to solve the problem of color distortion and loss of detail information in most dehazing algorithms, an end-to-end image dehazing network based on multi-scale feature enhancement is proposed. Firstly, the feature extraction enhancement module is used to capture the detailed information of hazy images and expand the receptive field. Secondly, the channel attention mechanism and pixel attention mechanism of the feature fusion enhancement module are used to dynamically adjust the weights of different channels and pixels. Thirdly, the context enhancement module is used to enhance the context semantic information, suppress redundant information, and obtain the haze density image with higher detail. Finally, our method removes haze, preserves image color, and ensures image details. The proposed method achieved a PSNR score of 33.74, SSIM scores of 0.9843 and LPIPS distance of 0.0040 on the SOTS-outdoor dataset. Compared with representative dehazing methods, it demonstrates better dehazing performance and proves the advantages of the proposed method on synthetic hazy images. Combined with dehazing experiments on real hazy images, the results show that our method can effectively improve dehazing performance while preserving more image details and achieving color fidelity.
ABSTRACT
The House-Tree-Person (HTP) sketch test is a psychological analysis technique designed to assess the mental health status of test subjects. Nowadays, there are mature methods for the recognition of depression using the HTP sketch test. However, existing works primarily rely on manual analysis of drawing features, which has the drawbacks of strong subjectivity and low automation. Only a small number of works automatically recognize depression using machine learning and deep learning methods, but their complex data preprocessing pipelines and multi-stage computational processes indicate a relatively low level of automation. To overcome the above issues, we present a novel deep learning-based one-stage approach for depression recognition in HTP sketches, which has a simple data preprocessing pipeline and calculation process with a high accuracy rate. In terms of data, we use a hand-drawn HTP sketch dataset, which contains drawings of normal people and patients with depression. In the model aspect, we design a novel network called Feature-Enhanced Bi-Level Attention Network (FBANet), which contains feature enhancement and bi-level attention modules. Due to the limited size of the collected data, transfer learning is employed, where the model is pre-trained on a large-scale sketch dataset and fine-tuned on the HTP sketch dataset. On the HTP sketch dataset, utilizing cross-validation, FBANet achieves a maximum accuracy of 99.07% on the validation dataset, with an average accuracy of 97.71%, outperforming traditional classification models and previous works. In summary, the proposed FBANet, after pre-training, demonstrates superior performance on the HTP sketch dataset and is expected to be a method for the auxiliary diagnosis of depression.