Important Feature Information Research Articles

A COVID-19 medical image classification algorithm based on Transformer

Coronavirus 2019 (COVID-19) is a new acute respiratory disease that has spread rapidly throughout the world. This paper proposes a novel deep learning network based on ResNet-50 merged transformer named RMT-Net. On the backbone of ResNet-50, it uses Transformer to capture long-distance feature information, adopts convolutional neural networks and depth-wise convolution to obtain local features, reduce the computational cost and acceleration the detection process. The RMT-Net includes four stage blocks to realize the feature extraction of different receptive fields. In the first three stages, the global self-attention method is adopted to capture the important feature information and construct the relationship between tokens. In the fourth stage, the residual blocks are used to extract the details of feature. Finally, a global average pooling layer and a fully connected layer perform classification tasks. Training, verification and testing are carried out on self-built datasets. The RMT-Net model is compared with ResNet-50, VGGNet-16, i-CapsNet and MGMADS-3. The experimental results show that the RMT-Net model has a Test_ acc of 97.65% on the X-ray image dataset, 99.12% on the CT image dataset, which both higher than the other four models. The size of RMT-Net model is only 38.5 M, and the detection speed of X-ray image and CT image is 5.46 ms and 4.12 ms per image, respectively. It is proved that the model can detect and classify COVID-19 with higher accuracy and efficiency.

A hyperautomative human behaviour recognition algorithm based on improved residual network

ABSTRACT When dealing with the mutual storage relationship of behavioral features in long time sequence video, the convolutional neural network is easy to miss important feature information. To solve the above problems, this paper proposes a super automatic algorithm combining nonlocal convolution and three-dimensional convolution neural network. The algorithm uses sparse sampling to segment the long time sequence video to reduce the amount of redundant information, and integrates non-local convolution into the residual neural network, thus forming a super automatic full variational - L1 algorithm. Experimental results show that the proposed method can significantly improve the efficiency of behavior recognition.

Multi-Channel Expression Recognition Network Based on Channel Weighting

Accurate expression interpretation occupies a huge proportion of human-to-human communication. The control of expressions can facilitate more convenient communication between people. Expression recognition technology has also been transformed from relatively mature laboratory-controlled research to natural scenes research. In this paper, we design a multi-channel attention network based on channel weighting for expression analysis in natural scenes. The network mainly consists of three parts: Multi-branch expression recognition feature extraction network, which combines residual network ResNet18 and ConvNeXt network ideas to improve feature extraction and uses adaptive feature fusion to build a complete network; Adaptive Channel Weighting, which designs adaptive weights in the auxiliary network for feature extraction, performs channel weighting, and highlights key information areas; and Attention module, which designs and modifies the spatial attention mechanism and increases the proportion of feature information to accelerate the acquisition of important expression feature information areas. The experimental results show that the proposed method achieves better recognition efficiency than existing algorithms on the dataset FER2013 under uncontrolled conditions, reaching 73.81%, and also achieves good recognition accuracy of 89.65% and 85.24% on the Oulu_CASIA and RAF-DB datasets, respectively.

Dual-Path Adversarial Generation Network for Super-Resolution Reconstruction of Remote Sensing Images

Satellite remote sensing images contain adequate ground object information, making them distinguishable from natural images. Due to the constraint hardware capability of the satellite remote sensing imaging system, coupled with the surrounding complex electromagnetic noise, harsh natural environment, and other factors, the quality of the acquired image may not be ideal for follow-up research to make suitable judgment. In order to obtain clearer images, we propose a dual-path adversarial generation network model algorithm that particularly improves the accuracy of the satellite remote sensing image super-resolution. This network involves a dual-path convolution operation in a generator structure, a feature mapping attention mechanism that first extracts important feature information from a low-resolution image, and an enhanced deep convolutional network to extract the deep feature information of the image. The deep feature information and the important feature information are then fused in the reconstruction layer. Furthermore, we also improve the algorithm structure of the loss function and discriminator to achieve a relatively optimal balance between the output image and the discriminator, so as to restore the super-resolution image closer to human perception. Our algorithm was validated on the public UCAS-AOD datasets, and the obtained results showed significantly improved performance compared to other methods, thus exhibiting a real advantage in supporting various image-related field applications such as navigation monitoring.

Frequency Domain Filtering Method for SSVEP-EEG Preprocessing

Steady-state visual evoked potential (SSVEP) signal collected from the scalp typically contains other types of electric signals, and it is important to remove these noise components from the actual signal by application of a pre-processing step for accurate analysis. High-pass or bandpass filtering of the SSVEP signal in the time domain is the most common pre-processing method. Because frequency is the most important feature information contained in the SSVEP signal, a technique for frequency-domain filtering of SSVEP was proposed here. In this method, the time-domain signal is extended to multi-dimensional signal by empirical mode decomposition (EMD), where each dimension represents a intrinsic mode function (IMF). The multi-dimensional signal is transformed to a frequency-domain signal by 2-D Fourier transform, the Gaussian high-pass filter function is constructed to perform high-pass filtering, and then the filtered signal is transformed to time domain by 2-D inverse Fourier transform. Finally, the filtered multi-dimensional intrinsic mode function is superimposed and averaged as the frequency-domain filtered signal. Compared with the time-domain filtering method, the experimental results revealed that frequency-domain filtering method effectively removed the baseline drift in signal and effectively suppressed the low-frequency interference component. This method was tested using data from public datasets and the results show that the proposed frequency-domain filtering method can significantly improve the feature recognition performance of canonical correlation analysis (CCA), filter bank canonical correlation analysis (FBCCA), and task-related component analysis (TRCA) methods. Thus, the results suggest that the application of frequency-domain filtering in the pre-processing stage allows improved noise removal. The proposed method extends SSVEP signal filtering from time-domain to frequency-domain, and the results suggest that this analysis tool significantly promotes the practical application of SSVEP systems.

RT-YOLO: A Residual Feature Fusion Triple Attention Network for Aerial Image Target Detection

In recent years, target detection of aerial images of unmanned aerial vehicle (UAV) has become one of the hottest topics. However, target detection of UAV aerial images often presents false detection and missed detection. We proposed a modified you only look once (YOLO) model to improve the problems arising in object detection in UAV aerial images: (1) A new residual structure is designed to improve the ability to extract features by enhancing the fusion of the inner features of the single layer. At the same time, triplet attention module is added to strengthen the connection between space and channel and better retain important feature information. (2) The feature information is enriched by improving the multi-scale feature pyramid structure and strengthening the feature fusion at different scales. (3) A new loss function is created and the diagonal penalty term of the anchor frame is introduced to improve the speed of training and the accuracy of reasoning. The proposed model is called residual feature fusion triple attention YOLO (RT-YOLO). Experiments showed that the mean average precision (mAP) of RT-YOLO is increased from 57.2% to 60.8% on the vehicle detection in aerial image (VEDAI) dataset, and the mAP is also increased by 1.7% on the remote sensing object detection (RSOD) dataset. The results show that the RT-YOLO outperforms other mainstream models in UAV aerial image object detection.

Identification of Defective Maize Seeds Using Hyperspectral Imaging Combined with Deep Learning

Seed quality affects crop yield and the quality of agricultural products, and traditional identification methods are time-consuming, complex, and irreversibly destructive. This study aims to establish a fast, non-destructive, and effective approach for defect detection in maize seeds based on hyperspectral imaging (HSI) technology combined with deep learning. Raw spectra collected from maize seeds (200 each healthy and worm-eaten) were pre-processed using detrending (DE) and multiple scattering correction (MSC) to highlight the spectral differences between samples. A convolutional neural network architecture (CNN-FES) based on a feature selection mechanism was proposed according to the importance of wavelength in the target classification task. The results show that the subset of 24 feature wavelengths selected by the proposed CNN-FES can capture important feature information in the spectral data more effectively than the conventional successive projections algorithm (SPA) and competitive adaptive reweighted sampling (CARS) algorithms. In addition, a convolutional neural network architecture (CNN-ATM) based on an attentional classification mechanism was designed for one-dimensional spectral data classification and compared with three commonly used machine learning methods, linear discriminant analysis (LDA), random forest (RF), and support vector machine (SVM). The results show that the classification performance of the designed CNN-ATM on the full wavelength does not differ much from the above three methods, and the classification accuracy is above 90% on both the training and test sets. Meanwhile, the accuracy, sensitivity, and specificity of CNN-ATM based on feature wavelength modeling can reach up to 97.50%, 98.28%, and 96.77% at the highest, respectively. The study shows that hyperspectral imaging-based defect detection of maize seed is feasible and effective, and the proposed method has great potential for the processing and analysis of complex hyperspectral data.

Litchi Detection in a Complex Natural Environment Using the YOLOv5-Litchi Model

Detecting litchis in a complex natural environment is important for yield estimation and provides reliable support to litchi-picking robots. This paper proposes an improved litchi detection model named YOLOv5-litchi for litchi detection in complex natural environments. First, we add a convolutional block attention module to each C3 module in the backbone of the network to enhance the ability of the network to extract important feature information. Second, we add a small-object detection layer to enable the model to locate smaller targets and enhance the detection performance of small targets. Third, the Mosaic-9 data augmentation in the network increases the diversity of datasets. Then, we accelerate the regression convergence process of the prediction box by replacing the target detection regression loss function with CIoU. Finally, we add weighted-boxes fusion to bring the prediction boxes closer to the target and reduce the missed detection. An experiment is carried out to verify the effectiveness of the improvement. The results of the study show that the mAP and recall of the YOLOv5-litchi model were improved by 12.9% and 15%, respectively, in comparison with those of the unimproved YOLOv5 network. The inference speed of the YOLOv5-litchi model to detect each picture is 25 ms, which is much better than that of Faster-RCNN and YOLOv4. Compared with the unimproved YOLOv5 network, the mAP of the YOLOv5-litchi model increased by 17.4% in the large visual scenes. The performance of the YOLOv5-litchi model for litchi detection is the best in five models. Therefore, YOLOv5-litchi achieved a good balance between speed, model size, and accuracy, which can meet the needs of litchi detection in agriculture and provides technical support for the yield estimation and litchi-picking robots.

A Video Description Model with Improved Attention Mechanism

Video description generation refers to the automatic generation of text descriptions of videos by computers, which belongs to the intersection of computer vision and natural language processing. Aiming at the problem that the traditional attention mechanism has insufficient ability to extract video features, the model is complex and the description quality is not high, this paper proposes a video description model with an improved attention mechanism. The model is based on the encoder-decoder structure, uses inception-v4 as the encoder to extract features, and introduces a lightweight coordinate attention module (CA) into the attention mechanism, which improves the feature extraction effect and reduces the model complexity, and sends the extracted important feature information into the decoder long short-term memory network (LSTM) to generate the description sentence corresponding to the video. The model is validated on the MSVD dataset using various evaluation metrics (BLEU, ROUGEL, CIDEr, METEOR). The experimental results show that the improved attention mechanism of the video description model proposed in this paper has better accuracy in different performance metrics and can further improve the performance of video description.

Attention-Based Scene Text Detection on Dual Feature Fusion.

The segmentation-based scene text detection algorithm has advantages in scene text detection scenarios with arbitrary shape and extreme aspect ratio, depending on its pixel-level description and fine post-processing. However, the insufficient use of semantic and spatial information in the network limits the classification and positioning capabilities of the network. Existing scene text detection methods have the problem of losing important feature information in the process of extracting features from each network layer. To solve this problem, the Attention-based Dual Feature Fusion Model (ADFM) is proposed. The Bi-directional Feature Fusion Pyramid Module (BFM) first adds stronger semantic information to the higher-resolution feature maps through a top-down process and then reduces the aliasing effects generated by the previous process through a bottom-up process to enhance the representation of multi-scale text semantic information. Meanwhile, a position-sensitive Spatial Attention Module (SAM) is introduced in the intermediate process of two-stage feature fusion. It focuses on the one feature map with the highest resolution and strongest semantic features generated in the top-down process and weighs the spatial position weight by the relevance of text features, thus improving the sensitivity of the text detection network to text regions. The effectiveness of each module of ADFM was verified by ablation experiments and the model was compared with recent scene text detection methods on several publicly available datasets.

LACN: A lightweight attention-guided ConvNeXt network for low-light image enhancement

Images captured under low-light conditions usually have poor visual quality, and hence greatly reduce the accuracy of subsequent tasks such as image segmentation and detection. In the low-light image enhancement task, noises in the dark areas are generally amplified while the images’ brightness is enhanced. It should be pointed out that many deep learning methods cannot effectively suppress the noise at this stage and capture important feature information. To address the above problem, this paper proposes a Lightweight Attention-guided ConvNeXt Network (LACN) for low-light image enhancement. A novel Attention ConvNeXt Module (ACM) is first proposed by introducing a parameter-free attention module (i.e. SimAM) into the ConvNeXt backbone network. Then, a nontrivial lightweight network LACN based on a multi-attention mechanism is established through stacking two ACMs and fusing their features. In what follows, an improved hybrid attention mechanism, Selective Kernel Attention Module (SKAM), is adopted to effectively extract both global and local information. Such a module realizes the evaluation of lighting conditions for the whole image and the adaptive adjustment of the receptive field. Finally, through the feature fusion module, the features of different stages are aggregated to improve the ability of network to retain color information. Numerous experiments on low-light image enhancement are implemented via comparison with other state-of-the-art methods. Experiments show that the proposed method significantly improves the brightness and contrast of low-illumination images, preserves color information, and suppresses the generation of noises after image brightening.

A novel fault diagnosis method of power cable based on convolutional probabilistic neural network with discrete wavelet transform and symmetrized dot pattern

To accurately diagnose the XLPE power cable insulation fault, this research proposed a novel hybrid algorithm combined with Convolutional Probabilistic Neural Network (CPNN) based on Discrete Wavelet Transform (DWT) and Symmetrized Dot Pattern (SDP) analysis. First, it built seven different power cable insulation defect models to measure partial discharge signals of power cable insulation faults. Then, a discrete wavelet transform was used for noise filtering. The time-domain partial discharge signal was directly converted into the point coordinate feature image of visual polar coordinates by SDP analyses. Finally, the feature image was trained and recognized by CPNN. After the important feature information of the feature-image was extracted by convolution layer and pooling layer operations, it is applied to the power cable insulation fault state diagnosis system based on the rapid learning and highly parallel computing of Probabilistic Neural Network (PNN). The experimental results proved that the method proposed in this research could accurately diagnose the power cable insulation fault type and the recognition accuracy is higher than 96%. The proposed method has a short detection time and high diagnostic accuracy. This proves that it can be applied to detect the power cable insulation fault type.

Research on Key Technologies of Super-Resolution Reconstruction of Medium and Long Wave Maritime Infrared Image

Complex illumination, solar flares and heavy smog on the sea surface have caused difficulties to accurately obtain high-quality imaging and multi-dimensional information of marine monitoring targets, such as oil spill, red tide and underwater vehicle wake. The principle of existing imaging mechanism is complex, and thus it is not practical to capture high-resolution infrared images efficiently. To combat this challenge by utilizing new infrared optical materials and single point diamond-turning technology, we designed and processed a simple, light and strong condensing ability medium and long wavelength infrared imaging optical system with large relative aperture, which can obtain high-quality infrared images. On top of this, with the training from a combination of infrared and visible light images, we also proposed a super-resolution network model, which is composed of a feature extraction layer, an information extraction block and a reconstruction block. The initial features of the input images are recognized in feature extraction layer. Next, to supply missing feature information and recover more details on infrared image extracted from a dense connection block, a feature mapping attention mechanism is introduced. Its main function is to transfer the important feature information of the visible light images in the information extraction block. Finally, the global feature information is integrated in the reconstruction block to reconstruct the high-resolution infrared image. We experimented our algorithm on both of the public Kaist datasets and self-collected datasets, and then compared it with several relevant algorithms. The results showed that our algorithm can significantly improve the reconstruction performance and reveal more detail information, and enhance the visual effect. Therefore, it brings excellent potential in dealing with the problem of low resolution of optical infrared imaging in complex marine environment.

PODD: A Dual-Task Detection for Greenhouse Extraction Based on Deep Learning

The rapid boom of the global population is causing more severe food supply problems. To deal with these problems, the agricultural greenhouse is an effective way to increase agricultural production within a limited space. To better guide agricultural activities and respond to future food crises, it is important to obtain both the agricultural greenhouse area and quantity distribution. In this study, a novel dual-task algorithm called Pixel-based and Object-based Dual-task Detection (PODD) that combines object detection and semantic segmentation is proposed to estimate the quantity and extract the area of agricultural greenhouses based on RGB remote sensing images. This algorithm obtains the quantity of agricultural greenhouses based on the improved You Only Look Once X (YOLOX) network structure, which is embedded with Convolutional Block Attention Module (CBAM) and Adaptive Spatial Feature Fusion (ASFF). The introduction of CBAM can make up for the lack of expression ability of its feature extraction layer to retain more important feature information. Adding the ASFF module can make full use of the features in different scales to increase the precision. This algorithm obtains the area of agricultural greenhouses based on the DeeplabV3+ neural network using ResNet-101 as a feature extraction network, which not only effectively reduces hole and plaque issues but also extracts edge details. Experimental results show that the mAP and F1-score of the improved YOLOX network reach 97.65% and 97.50%, 1.50% and 2.59% higher than the original YOLOX solution. At the same time, the accuracy and mIoU of the DeeplabV3+ network reach 99.2% and 95.8%, 0.5% and 2.5% higher than the UNet solution. All of the metrics in the dual-task algorithm reach 95% and even higher. Proving that the PODD algorithm could be useful for agricultural greenhouse automatic extraction (both quantity and area) in large areas to guide agricultural policymaking.

Scale-Space Feature Recalibration Network for Single Image Deraining.

Computer vision technology is increasingly being used in areas such as intelligent security and autonomous driving. Users need accurate and reliable visual information, but the images obtained under severe weather conditions are often disturbed by rainy weather, causing image scenes to look blurry. Many current single image deraining algorithms achieve good performance but have limitations in retaining detailed image information. In this paper, we design a Scale-space Feature Recalibration Network (SFR-Net) for single image deraining. The proposed network improves the image feature extraction and characterization capability of a Multi-scale Extraction Recalibration Block (MERB) using dilated convolution with different convolution kernel sizes, which results in rich multi-scale rain streaks features. In addition, we develop a Subspace Coordinated Attention Mechanism (SCAM) and embed it into MERB, which combines coordinated attention recalibration and a subspace attention mechanism to recalibrate the rain streaks feature information learned from the feature extraction phase and eliminate redundant feature information to enhance the transfer of important feature information. Meanwhile, the overall SFR-Net structure uses dense connection and cross-layer feature fusion to repeatedly utilize the feature maps, thus enhancing the understanding of the network and avoiding gradient disappearance. Through extensive experiments on synthetic and real datasets, the proposed method outperforms the recent state-of-the-art deraining algorithms in terms of both the rain removal effect and the preservation of image detail information.

Multi-Level Cycle-Consistent Adversarial Networks with Attention Mechanism for Face Sketch-Photo Synthesis

The synthesis between face sketches and face photos has important application values in law enforcement and digital entertainment. In cases of a lack of paired sketch-photo data, this paper proposes an unsupervised model to solve the problems of missing key facial details and a lack of realism in the synthesized images of existing methods. The model is built on the CycleGAN architecture. To retain more semantic information in the target domain, a multi-scale feature extraction module is inserted before the generator. In addition, the convolutional block attention module is introduced into the generator to enhance the ability of the model to extract important feature information. Via CBAM, the model improves the quality of the converted image and reduces the artifacts caused by image background interference. Next, in order to preserve more identity information in the generated photo, this paper constructs the multi-level cycle consistency loss function. Qualitative experiments on CUFS and CUFSF public datasets show that the facial details and edge structures synthesized by our model are clearer and more realistic. Meanwhile the performance indexes of structural similarity and peak signal-to-noise ratio in quantitative experiments are also significantly improved compared with other methods.

COVID-19 CT image segmentation method based on swin transformer.

Owing to its significant contagion and mutation, the new crown pneumonia epidemic has caused more than 520 million infections worldwide and has brought irreversible effects on the society. Computed tomography (CT) images can clearly demonstrate lung lesions of patients. This study used deep learning techniques to assist doctors in the screening and quantitative analysis of this disease. Consequently, this study will help to improve the diagnostic efficiency and reduce the risk of infection. In this study, we propose a new method to improve U-Net for lesion segmentation in the chest CT images of COVID-19 patients. 750 annotated chest CT images of 150 patients diagnosed with COVID-19 were selected to classify, identify, and segment the background area, lung area, ground glass opacity, and lung parenchyma. First, to address the problem of a loss of lesion detail during down sampling, we replaced part of the convolution operation with atrous convolution in the encoder structure of the segmentation network and employed convolutional block attention module (CBAM) to enhance the weighting of important feature information. Second, the Swin Transformer structure is introduced in the last layer of the encoder to reduce the number of parameters and improve network performance. We used the CC-CCII lesion segmentation dataset for training and validation of the model effectiveness. The results of ablation experiments demonstrate that this method achieved significant performance gain, in which the mean pixel accuracy is 87.62%, mean intersection over union is 80.6%, and dice similarity coefficient is 88.27%. Further, we verified that this model achieved superior performance in comparison to other models. Thus, the method proposed herein can better assist doctors in evaluating and analyzing the condition of COVID-19 patients.

Research on Pedestrian Detection and DeepSort Tracking in Front of Intelligent Vehicle Based on Deep Learning

In order to improve the tracking failure caused by small-target pedestrians and partially blocked pedestrians in dense crowds in complex environments, a pedestrian target detection and tracking method for an intelligent vehicle was proposed based on deep learning. On the basis of the YOLO detection model, the channel attention module and spatial attention module were introduced and were joined to the back of the backbone network Darknet-53 in order to achieve weight amplification of important feature information in channel and space dimensions and improve the representation ability of the model for important feature information. Based on the improved YOLO network, the flow of the DeepSort pedestrian tracking method was designed and the Kalman filter algorithm was used to estimate the pedestrian motion state. The Mahalanobis distance and apparent feature were used to calculate the similarity between the detection frame and the predicted pedestrian trajectory; the Hungarian algorithm was used to achieve the optimal matching of pedestrian targets. Finally, the improved YOLO pedestrian detection model and the DeepSort pedestrian tracking method were verified in the same experimental environment. The verification results showed that the improved model can improve the detection accuracy of small-target pedestrians, effectively deal with the problem of target occlusion, reduce the rate of missed detection and false detection of pedestrian targets, and improve the tracking failure caused by occlusion.

HRNet Encoder and Dual-Branch Decoder Framework-Based Scene Text Recognition Model

Scene text recognition (STR) is designed to automatically recognize the text content in natural scenes. Different from regular document text, text in natural scenes has the characteristics of irregular shapes, complex background, and distorted and blurred contents, which makes STR challenging. To solve the problems of STR for distorted, blurred, and low-resolution texts in natural scenes, this paper proposes a HRNet encoder and dual-branch decoder framework-based STR model. The model mainly consists of an encoder module and a dual-branch decoder module composed of a super-resolution branch and a recognition branch in parallel. In the encoder module, the HRNet is adopted to realize the cross-parallel aggregation representation with multiple resolutions during feature extraction and then outputs four kinds of feature maps with different resolutions. Moreover, the supervised attention module is used to strengthen the learning of the important feature information. In the decoder module, the dual-branch structure is adopted, in which the super-resolution branch takes the feature maps with the highest resolution obtained in the encoder module as input and restores images by upsampling through transposed convolution. The four kinds of feature maps with different resolutions are fused through independent transposed convolution layers for multiscale fusion in the recognition branch and then inputted into the attention-based decoder for text recognition. To improve the accuracy of text recognition, the feature extraction effect of the encoder module is together supervised by the super-resolution branch loss and the recognition branch loss. In addition, the super-resolution branch is only used for training and is abandoned during testing to reduce the complexity of the model. The proposed model is trained on Synth90K and SynthText datasets and tested on seven natural scene datasets. Compared with classical models such as ASTER, TextSR, and SCGAN, the recognition accuracy of the proposed model is improved and better recognition results can be achieved on irregular and blurred datasets such as IC15, SVTP, and CUTE80.

Improving YOLOv4-Tiny’s Construction Machinery and Material Identification Method by Incorporating Attention Mechanism

To facilitate the development of intelligent unmanned loaders and improve the recognition accuracy of loaders in complex scenes, we propose a construction machinery and material target detection algorithm incorporating an attention mechanism (AM) to improve YOLOv4-Tiny. First, to ensure the robustness of the proposed algorithm, we adopt style migration and sliding window segmentation to increase the underlying dataset’s diversity. Second, to address the problem that YOLOv4-Tiny’s (the base network) framework only adopts a layer-by-layer connection form, which demonstrates an insufficient feature extraction ability, we adopt a multilayer cascaded residual module to deeply connect low- and high-level information. Finally, to filter redundant feature information and make the proposed algorithm focus more on important feature information, a channel AM is added to the base network to perform a secondary screening of feature information in the region of interest, which effectively improves the detection accuracy. In addition, to achieve small-scale object detection, a multiscale feature pyramid network structure is employed in the prediction module of the proposed algorithm to output two prediction networks with different scale sizes. The experimental results show that, compared with the traditional network structure, the proposed algorithm fully incorporates the advantages of residual networks and AM, which effectively improves its feature extraction ability and recognition accuracy of targets at different scales. The final proposed algorithm exhibits the features of high recognition accuracy and fast recognition speed, with mean average precision and detection speed reaching 96.82% and 134.4 fps, respectively.