Target Feature Extraction Research Articles

Novel feature extraction of underwater targets by encoding hydro-acoustic signatures as image

Underwater vessel-radiated acoustical noise (UVRAN) is a major factor for classification in the sea by the SONAR. Due to unsteady and complex maritime ambient, analyzing underwater sound signals is a challenging issue that has lately received attention in the marine field. In the conventional feature extraction methods, to reduce the effect of ocean noise, the de-noising procedure is performed before complexity measurement by mode decomposition techniques. Based on this, we propose a novel insight for the first time to distinguish the objects which made the underwater noises as the hydro-acoustic signature, using a signals-to-image conversion without noise removal. After pre-processing, the spectral amplitude mean difference function is encoded into an image using Gramian angular field (GAF) technique. Subsequently, image texture analysis is performed in which GAF images are subjected to the gray-level co-occurrence matrix (GLCM). Finally, the second-order image statistic (i.e., 2-D permutation entropy) is calculated. Compared with other methods, results demonstrate that the proposed method has a high degree of separation and stability between the various kinds of underwater targets, suggesting that the methodology is superior to the existing methods. Moreover, our model is robust to noise. The approach perhaps opens an alternative path for UVRAN discrimination.

Multimodal Fine-Grained Transformer Model for Pest Recognition

Deep learning has shown great potential in smart agriculture, especially in the field of pest recognition. However, existing methods require large datasets and do not exploit the semantic associations between multimodal data. To address these problems, this paper proposes a multimodal fine-grained transformer (MMFGT) model, a novel pest recognition method that improves three aspects of transformer architecture to meet the needs of few-shot pest recognition. On the one hand, the MMFGT uses self-supervised learning to extend the transformer structure to extract target features using contrastive learning to reduce the reliance on data volume. On the other hand, fine-grained recognition is integrated into the MMFGT to focus attention on finely differentiated areas of pest images to improve recognition accuracy. In addition, the MMFGT further improves the performance in pest recognition by using the joint multimodal information from the pest’s image and natural language description. Extensive experimental results demonstrate that the MMFGT obtains more competitive results compared to other excellent models, such as ResNet, ViT, SwinT, DINO, and EsViT, in pest recognition tasks, with recognition accuracy up to 98.12% and achieving 5.92% higher accuracy compared to the state-of-the-art DINO method for the baseline.

Real-time signal processing method of ultra-wideband fuze using LSTM

The extraction of target features from the ultra-wideband (UWB) fuze echo signal and target recognition and distance determination are the key to controlling the precise detonation of the fuze. In this paper, a real-time signal processing method of UWB fuze based on Long Short-Term Memory (LSTM) network is proposed. The time domain and frequency domain features of echo signal are recognized through a small-scale LSTM network. Matlab is used to build the network model and simulation results show that the method can effectively avoid noise interference, correctly recognize the target signal, and avoid false alarm.

Small Target Detection Method Based on Low-Rank Sparse Matrix Factorization for Side-Scan Sonar Images

Target detection in side-scan sonar images plays a significant role in ocean engineering. However, the target images are usually severely interfered by the complex background and strong environmental noise, which makes it difficult to extract robust features from small targets and makes the target detection task quite challenging. In this paper, a novel small target detection method in sonar images is proposed based on the low-rank sparse matrix factorization. Initially, the side-scan sonar images are preprocessed so as to highlight the individual differences of the target. Then, the problems of target feature extraction and noise removal are characterized as the problem of matrix decomposition. An improved Robust Principal Component Analysis algorithm is used to extract target information, and the fast proximal gradient method is used to optimize the solution. The original sonar image is reconstructed into the low-rank background matrix, the sparse target matrix, and the noise matrix. Eventually, a morphological operation is used to filter out the noise and refine the target edges in the target matrix for improving the accuracy of target detection. Experimental results show that the proposed method not only achieves better detection performance in comparison to the conventional baseline algorithms but also performs robustly in various signal-to-clutter ratio conditions.

The clinical significance of computed tomography texture features of renal cell carcinoma in predicting pathological T1-3 staging.

Precise T staging is an important prerequisite for the treatment decisions of patients with renal cell carcinoma (RCC). We aimed to predict the pathological T1-3 staging of RCC with an automatic multiclass T staging prediction mode. We retrospectively enrolled 100 consecutive patients with pathologically proven RCC that was newly diagnosed and untreated from Sun Yat-sen University Cancer Center and randomly split these patients into a training set (70%) and an internal testing set (30%). We enrolled additional 29 patients with pathologically proven RCC from The Third Affiliated Hospital of Shenzhen University as the external testing set. We used the training set data to establish a prediction model for pathological T1-3 staging of RCC and validated the effect of the training model using the internal and external testing sets. Quantitative decomposition of the prediction model was conducted to explore the contribution of each extracted feature. The computed tomography (CT) images of 100 patients (37, 29, and 34 patients with T1, T2, and T3 staging, respectively, according to the eighth tumor-node-metastasis staging system) were used to establish the prediction model for T staging using delineation of the target area, image segmentation, and feature extraction. The micro area under the curve (AUC) and macro-AUC of the model were 0.90 [95% confidence interval (CI): 0.84-1.00] and 0.91 (95% CI: 0.86-1.00), respectively. In terms of validation with the external testing set, the micro-AUC and macro-AUC were 0.72 (95% CI: 0.66-0.84) and 0.78 (95% CI: 0.69-0.88), respectively. Our prediction model showed good performance in predicting the pathological T1-3 staging of RCC.

Low sampling high quality image reconstruction and segmentation based on array network ghost imaging.

High-quality imaging under low sampling time is an important step in the practical application of computational ghost imaging (CGI). At present, the combination of CGI and deep learning has achieved ideal results. However, as far as we know, most researchers focus on one single pixel CGI based on deep learning, and the combination of array detection CGI and deep learning with higher imaging performance has not been mentioned. In this work, we propose a novel multi-task CGI detection method based on deep learning and array detector, which can directly extract target features from one-dimensional bucket detection signals at low sampling times, especially output high-quality reconstruction and image-free segmentation results at the same time. And this method can realize fast light field modulation of modulation devices such as digital micromirror device to improve the imaging efficiency by binarizing the trained floating-point spatial light field and fine-tuning the network. Meanwhile, the problem of partial information loss in the reconstructed image due to the detection unit gap in the array detector has also been solved. Simulation and experimental results show that our method can simultaneously obtain high-quality reconstructed and segmented images at sampling rate of 0.78 %. Even when the signal-to-noise ratio of the bucket signal is 15 dB, the details of the output image are still clear. This method helps to improve the applicability of CGI and can be applied to resource-constrained multi-task detection scenarios such as real-time detection, semantic segmentation, and object recognition.

Target recognition using rotating ultrasonic sensor for an amphibious ROV

Recognition and tracking of nearby obstacle targets in the inshore environment are necessary for the driving of amphibious remotely operated vehicles (ROV). In this study, the underwater target recognition and tracking in front of the ROV was conducted with a cost-effective waterproof rotating ultrasonic sensing system. An empirical detection model was developed to establish the detection characteristics of the ultrasonic sensor in the turbid underwater environment. The feature extraction of the underwater targets was performed with the density-based spatial clustering of applications with noise. The maximum relative error of targets geometric feature recognition is 3.85%. The target tracking algorithm was designed based on the Kalman filter. The rotating ultrasonic sensing system was further applied and validated on a full size amphibious tracked ROV in the turbid underwater environment. The experimental results show that the maximum recognition error during the straight movement is less than 5% after Kalman filter processing. The validity of the DBSCAN and Kalman filtering algorithms is verified during the driving of the amphibious ROV.

MO-SOD: Micro-Oxidation Small Object Detection Model for Oxygen-Free Copper Surfaces Based on Microscopic Imaging System.

Micro-oxidation is a fatal problem for some precision oxygen-free copper materials, and it is difficult to detect with the naked eyes. However, manual inspection using microscope equipment is expensive, subjective, and time-consuming. The automatic high-definition micrograph system equipped with micro-oxidation detection algorithm can detect more quickly, efficiently, and accurately. In this study, a micro-oxidation small objection detection model, MO-SOD, is proposed to detect the oxidation degree on oxygen-free copper surface based on microimaging system. This model is developed for rapid detection on the robot platform combined with high-definition microphotography system. The proposed MO-SOD model consists of three modules: small target feature extraction layer, key small object attention pyramid integration layer, and anchor-free decoupling detector. The small object feature extraction layer focuses on the local features of small object to improve the perception of micro-oxidation spots and also takes the global features into account to reduce the impact of noisy background on feature extraction. Key small object attention pyramid integration block couples key small object feature attention and pyramid to detect the micro-oxidation spots in the image. The performance of MO-SOD model is further improved by combining the anchor-free decoupling detector. In addition, the loss function is improved to combine CIOU loss and focal loss to achieve effective micro-oxidation detection. The MO-SOD model is trained and tested from three oxidation levels in an oxygen-free copper surface microscope image data set. The test results show that the average accuracy (mAP) of MO-SOD model is 82.96%, which is superior to other most advanced detectors.

Research on steel rail surface defects detection based on improved YOLOv4 network.

The surface images of steel rails are extremely difficult to detect and recognize due to the presence of interference such as light changes and texture background clutter during the acquisition process. To improve the accuracy of railway defects detection, a deep learning algorithm is proposed to detect the rail defects. Aiming at the problems of inconspicuous rail defects edges, small size and background texture interference, the rail region extraction, improved Retinex image enhancement, background modeling difference, and threshold segmentation are performed sequentially to obtain the segmentation map of defects. For the classification of defects, Res2Net and CBAM attention mechanism are introduced to improve the receptive field and small target position weights. The bottom-up path enhancement structure is removed from the PANet structure to reduce the parameter redundancy and enhance the feature extraction of small targets. The results show the average accuracy of rail defects detection reaches 92.68%, the recall rate reaches 92.33%, and the average detection time reaches an average of 0.068 s per image, which can meet the real-time of rail defects detection. Comparing the improved method with the mainstream target detection algorithms such as Faster RCNN, SSD, YOLOv3 and other algorithms, the improved YOLOv4 has excellent comprehensive performance for rail defects detection, the improved YOLOv4 model obviously better than several others in P r , R c , and F1 value, and can be well-applied to rail defect detection projects.

KPE-YOLOv5: An Improved Small Target Detection Algorithm Based on YOLOv5

At present, the existing methods have many limitations in small target detection, such as low accuracy, a high rate of false detection, and missed detection. This paper proposes the KPE-YOLOv5 algorithm aiming to improve the ability of small target detection. The algorithm has three improvements based on the YOLOv5 algorithm. Firstly, it achieves more accurate size of anchor-boxes for small targets by K-means++ clustering technology. Secondly, the scSE (spatial and channel compression and excitation) attention module is integrated into the new algorithm to encourage the backbone network to pay greater attention to the feature information of small targets. Finally, the capability of small target feature extraction is improved by increasing the small target detection layer, which also increases the detection accuracy of small targets. We evaluate KPE-YOLOv5 on the VisDrone-2020 dataset and compare performance with YOLOv5. The results show that KPE-YOLOv5 improves the detection mAP by 5.3% and increases the P by 7%. The KPE-YOLOv5 algorithm has better detection outcome than YOLOv5 for small target detection.

SAR image near-shore ship target detection method in complex background

ABSTRACT Due to background clutter in synthetic aperture radar (SAR) images, the detection of dense ship targets suffers from a low detection rate, high false alarm rate, and high missed detection rate. To address this issue, an FSM-DFF-YOLOv5+Confluence algorithm is proposed in this paper for the detection of near-shore ship targets in SAR images with complex backgrounds. First, based on the YOLOv5 target detection algorithm, two improvements are made in the feature extraction network: feature refinement and multi-feature fusion; in the feature extraction network, deformable convolutional neural networks are adopted to change the position of the target sampling points of the convolution to improve the feature extraction capability of the target and the detection rate of ship targets in SAR images with a complex background; in the multi-feature fusion network structure, cascading and parallel pyramids are used in the multi-feature fusion network to realize feature fusion at different levels; the visual perceptual field of feature extraction is expanded by using null convolution to enhance the adaptability of the network to detect near-shore multi-scale ship targets with complex backgrounds and reduce the false alarm rate of ship target detection in SAR images with complex environments. In this way, the DFF-YOLOv5 near-shore ship target detection algorithm is established. Meanwhile, to address the problem of missed detection in near-shore dense ship target detection, this paper adds rectangular convolution kernels to the convolution of the feature extraction network to better realize the feature extraction of dense ship targets in SAR images with complex backgrounds. Besides, the Confluence algorithm instead of non-maximum suppression is used in the prediction stage. Through experiments on the constructed complex background near-shore ship detection dataset, it is indicated that the average accuracy of the FSM-DFF-YOLOv5+Confluence detection algorithm reaches 88.96%, and the recall rate reaches 88.80%.

A novel method for real-time ATR system of AUV based on Attention-MobileNetV3 network and pixel correction algorithm

This paper proposes an Automatic Target Recognition (ATR) system based on the Autonomous Underwater Vehicle (AUV) equipped with side scan sonar for real-time detection and positioning of underwater targets. The system introduces a pre-processing algorithm for the original Side Scan Sonar (SSS) image, which performs pixel-level correction of the image according to the slant range setting of the SSS sensor and the height from the seabed to eliminate the influence of target geometric distortion on the subsequent recognition algorithm and improve the recognition accuracy. At the same time, the attitude angle information of the AUV is obtained to correct the target position, making each pixel of the pre-processed image has underwater positioning (latitude and longitude) information. Since SSS images are seriously disturbed by noise, in order to accurately extract target features in real time, on the basis of selecting the lightweight network MobileNetV3, we introduce attention mechanism and integrated learning to meet the requirements of underwater target recognition accuracy and real-time performance. The multi-classifier ensemble prediction results further reduce the mis-identification rate of submarine targets. The ATR system fuses the prediction results and the submarine target position and sends it to the AUV main control system to guide the AUV how to navigate and plan path at the subsequent. The experimental results show that, we validate that the proposed method matches the state of the art in segmentation accuracy, the ATR system proposed in this paper can be embedded in the underwater unmanned platform, and achieve accurate underwater target detection and positioning, The real-time recognition speed of each picture is about 0.058 s.

Synthetic aperture radar target recognition via deep attention convolutional network assisted by multiscale residual despeckling network

Owing to the inherent characteristics of synthetic aperture radar (SAR) imaging system based on the coherent imaging modality, SAR images are inevitably corrupted by speckle noise, thereby affecting SAR target recognition accuracy. We explore the impact of speckle noise on SAR target recognition performance. Subsequently, we propose a deep attention convolutional network assisted by a multiscale residual despeckling network to improve SAR target recognition accuracy under speckle corruption. Noise information is first learned via a despeckling subnetwork consisting of dual-branch multiscale feature extraction, feature fusion, and adaptive feature channel selection. Then a classification subnetwork with a cross-dimension interaction attention mechanism is designed to realize feature extraction and identity reasoning of SAR targets. Experimental results on the benchmark moving and stationary target acquisition and recognition dataset demonstrate the effectiveness and superiority of the proposed method.

Regression-Selective Feature-Adaptive Tracker for Visual Object Tracking

As a challenging visual task, visual object tracking has recently been composed of the classification and regression subtasks. The anchor-free regression network gets rid of the dependence on the anchors, but the redundant range makes it usually regress some samples involving non-target information. Evenly dividing a target by the regular receptive field often causes ambiguous target localization. To address these issues, we propose a regression-selective feature-adaptive tracker (<i>RSFA</i>), where the regression-selective subnetwork can not only free the regression task from anchors, but can also select more effective regression samples using the refined criterion. The proposed feature-adaptive strategy concentrates the classification subnetwork on target feature extraction via adaptively modifying the receptive field, and the attached centrality branch offers a correction for target localization by exploiting the spatial information. Additionally, the designed online update mechanism realizes the tracker&#x0027;s online optimization, improving robustness against target deformation. Extensive experiments are conducted on challenging benchmarks, including GOT10K, OTB2015, UAV123, NFS, VOT2018, VOT2019 and VOT2020-ST. Our tracker achieves satisfactory tracking results, and the evaluations of its tracking performance rank first or second in comparison with the state-of-the-art tracking algorithms.

High-resolution imaging and micro-motion feature extraction of space multiple targets

In radar applications, different micro-motion forms can be used as the basis of target recognition. However, the radar echo signal of multiple spatial targets is overlapping in the time-frequency and time-range domains, which increases the difficulty of micro-motion feature extraction. In this paper, a high-resolution imaging and micro-motion feature extraction framework based on a multiple joint-domain radar tool is proposed to address this mixed signal. First, an accurate and suitable micro-motion model of cone-shaped space multiple targets is built. Then, the two-dimensional adaptive regularized smoothed L0 norm algorithm (2D AReSL0) based on sparse reconstruction is formulated to directly reconstruct the inverse synthetic aperture radar (ISAR) image. Additionally, the range-frequency-time radar data cube can be extracted from the ISAR movie by the CLEAN algorithm. To solve the scattering point association problem in the radar data cube, a 3D segmentation Viterbi algorithm is designed to extract the micro-motion features. Finally, simulation and experiment results demonstrate the effectiveness of the proposed framework.

An FPGA-implemented Method for Real-time Multi-dimensional Feature Extraction of Sequence Image Targets

An FPGA-implemented Method for Real-time Multi-dimensional Feature Extraction of Sequence Image Targets

A peanut and weed detection model used in fields based on BEM-YOLOv7-tiny.

Due to the different weed characteristics in peanut fields at different weeding periods, there is an urgent need to study a general model of peanut and weed detection and identification applicable to different weeding periods in order to adapt to the development of mechanical intelligent weeding in fields. To this end, we propose a BEM-YOLOv7-tiny target detection model for peanuts and weeds identification and localization at different weeding periods to achieve mechanical intelligent weeding in peanut fields at different weeding periods. The ECA and MHSA modules were used to enhance the extraction of target features and the focus on predicted targets, respectively, the BiFPN module was used to enhance the feature transfer between network layers, and the SIoU loss function was used to increase the convergence speed and efficiency of model training and to improve the detection performance of the model in the field. The experimental results showed that the precision, recall, mAP and F1 values of the BEM-YOLOv7-tiny model were improved by 1.6%, 4.9%, 4.4% and 3.2% for weed targets and 1.0%, 2.4%, 2.2% and 1.7% for all targets compared with the original YOLOv7-tiny. The experimental results of positioning error show that the peanut positioning offset error detected by BEM-YOLOv7-tiny is less than 16 pixels, and the detection speed is 33.8 f/s, which meets the requirements of real-time seedling grass detection and positioning in the field. It provides preliminary technical support for intelligent mechanical weeding in peanut fields at different stages.

Byte-Level Function-Associated Method for Malware Detection

The byte stream is widely used in malware detection due to its independence of reverse engineering. However, existing methods based on the byte stream implement an indiscriminate feature extraction strategy, which ignores the byte function difference in different segments and fails to achieve targeted feature extraction for various byte semantic representation modes, resulting in byte semantic confusion. To address this issue, an enhanced adversarial byte function associated method for malware backdoor attack is proposed in this paper by categorizing various function bytes into three functions involving structure, code, and data. The Minhash algorithm, grayscale mapping, and state transition probability statistics are then used to capture byte semantics from the perspectives of text signature, spatial structure, and statistical aspects, respectively, to increase the accuracy of byte semantic representation. Finally, the three-channel malware feature image is constructed based on different function byte semantics, and a convolutional neural network is applied for detection. Experiments on multiple data sets from 2018 to 2021 show that the method can effectively combine byte functions to achieve targeted feature extraction, avoid byte semantic confusion, and improve the accuracy of malware detection.

A Fast Detection Method Using Anisotropic Guidance for Infrared Small Target Under Complex Scenes

The development of intelligent control and visual navigation, along with the ongoing increase in imaging distance, has led to realistic demands for weak and small target detection. However, figuring out how to reduce the interference of various complex factors while achieving a high detection rate and low false alarm rate for real-time infrared small target detection is still a significant challenge. We propose a computationally simple single-frame infrared small target fast detection method applicable to complex scenes. First, a novel anisotropy filter bank suitable for small target feature extraction is constructed and modified with a point spread function. This makes it possible to map the salient features and spatial features of the target in the spatial domain using the imaging mechanism. We then combined two sets of feature parameters using the filtered features, including isotropy measure and directional energy factors, to enable the detection of small targets. Experimental findings showed that the algorithm performs well in scenarios involving complicated clouds, urban ground, and vegetation. The software also achieved excellent real-time performance because of its straightforward computational design.

Harmonic Content-Based Protection Method for Microgrids via 1-Dimensional Recursive Median Filtering Algorithm

Microgrids (MGs) offers grid-connected (GC) and islanded (ID) operational modes. However, these dynamic modes of operation pose different microgrid protection challenges. This paper presents a new protection method for MGs based on a discrete one-dimensional recursive Median filter (1-DRMF). In the first step, the 1-DRMF is applied on a measured current signal on every single phase individually for targeted feature extraction. Then, the median filter deviation (MFD) and the selected harmonic distortion (SHD) are computed from the current signals of all phases independently. In the second step, the upsurges in the MFD and the SHD of all phases are cross-checked with the pre-established threshold value of 0.3 to identify and categorize fault incidents. Finally, the directional properties of three-phase (3-p) reactive energy are employed in order to pinpoint the faulty line section (LS). Many simulations were executed on MATLAB/Simulink to validate the sustainable performance of the established method. Results prove that the scheme can detect, classify, and locate the solid and high impedance faults (HIF) in the GC as well as the ID modes under radial and meshed scenarios.