YOLO-Extract: Improved YOLOv5 for Aircraft Object Detection in Remote Sensing Images

Traditional semantic segmentation methods have problems such as poor multi-scale feature extraction ability, weak lightweight backbone network feature extraction ability, lack of effective fusion of context information, resulting in edge segmentation errors and feature discontinuity. In this paper, a novel semantic segmentation model based on multi-layer information fusion and dual convolutional attention mechanism is proposed. In this method, SegFormer network is used as the backbone network, and multi-scale features of encoder output are fused with overlapping features. The feature extraction subnetwork is optimized by constructing the object region enhancement module, and the intermediate feature map is refined adaptively in each convolutional block of the deep network, so as to strengthen the fine extraction of multi-dimensional feature information of complex images. Dual convolutional attention module is used to fusion high-level semantic information to avoid the loss of feature information caused by up-sampling operation and the influence of introducing noise, and refine the effect of target edge segmentation. At the same time, the feature pyramid grid is proposed to process the overlapping features, obtain the context information of different scales, and enhance the semantic expression of features. Finally, the features processed by the feature pyramid grid module are combined to improve the segmentation effect. The experimental results on the public data set show that the proposed method has better performance than the existing methods, and has better segmentation effect on the object edge in the scene.

We propose a novel machine learning model for classification problems, Deep Twin Support Vector Machine (DTWSVM), which combines TWSVM with deep learning ideas. TWSVM is a successful algorithm for classification problems which seeks two nonparallel hyper planes to make each hyper plane close to one class and far from the other as much as possible. And Deep Learning (DL) models have shown good ability in feature extraction and dimension reduction by constructing multi-layer network. Since the feature extraction in DL can reduce feature dimension while maintain the main information of original inputs, we consider constructing a three layer network which contains input layer, hidden layer and output layer. We use two TWSVMs in the hidden layer to extract features based on the projection principle which is derived from Multi-Layer Perceptron (MLP). The two TWSVMs will get four hyper planes by solving four convex quadratic programs. A new dataset which consists of the extracted features with four feature dimensions is produced from the hidden layer and then we can input it to the main TWSVM of the output layer to make final prediction. Similar as TWSVM, we design linear DTWSVM and nonlinear DTWSVM which have been proved to be very effective in classification problems. In numerical experiments, we have obtained 100% prediction accuracy for several datasets which is state-of-the-art performance absolutely!

In recent years, convolutional neural networks (CNNs) have been successfully applied to inverse synthetic aperture radar (ISAR) sparse imaging because of their powerful ability in feature extraction. However, these CNNs only adopt single path feed-forward architectures and lack paths for directly transmitting original feature representations (OFRs) in shallow layers to reconstruction layers, which limits the complete reconstruction of target shape due to the underutilization of the OFRs that are efficient for recovering target details. Later, fully CNN (FCNN) introduces several skip connections (SKs) to establish the additional ways for directly passing the OFRs to the reconstruction layers. Nevertheless, the transmitted OFRs inevitably include the feature information of artifacts, which usually results the appearance of artifacts in final reconstructed target image. To address this issue, we introduce the gate units to FCNN, and refer to the improved FCNN as G-FCNN. Furthermore, the learnable gate units weight the OFRs transmitted by SKs and autonomously decide how many OFRs are transmitted further. To circumvent the shortage of the real data available for network training, we utilize the transfer learning strategy to guarantee a good performance of the G-FCNN. The imaging results of real data show that the G-FCNN-based imaging method is superior to the existing CNN-based imaging methods.

Misophonia is a condition characterized by an abnormal emotional response to specific sounds, such as eating, breathing, and clock ticking noises. Sound classification for misophonia is an important area of research since it can benefit in the development of interventions and therapies for individuals affected by the condition. In the area of sound classification, deep learning algorithms such as Convolutional Neural Networks (CNNs) have achieved a high accuracy performance and proved their ability in feature extraction and modeling. Recently, transformer models have surpassed CNNs as the dominant technology in the field of audio classification. In this paper, a transformer-based deep learning algorithm is proposed to automatically identify trigger sounds and the characterization of these sounds using acoustic features. The experimental results demonstrate that the proposed algorithm can classify trigger sounds with high accuracy and specificity. These findings provide a foundation for future research on the development of interventions and therapies for misophonia.

In a point cloud registration task, the feature extraction ability of the network directly affects the final registration effect. Previous work has tended to ignore the local information of point clouds and the context information between two point clouds in the feature extraction. To overcome the poor feature extraction ability and lack of robustness of existing registration algorithms, this paper, inspired by the graph attention network, proposes a learning-based point cloud registration framework that uses the existing network to extract point cloud features. We call it CPCRNet. Furthermore, a Capture module based on graph attention (GAT) is proposed to re-integrate the point cloud context information. Experimental results indicate that Capture, the core feature extraction module in the network, has a significant impact on the registration effect, with a mean square error, root mean square error, and mean absolute error between the ground truth and predicted values lower than those of competing methods. Moreover, experimental results for ModelNet40 and ShapeNet Part datasets show that the errors are not significantly changed by noise, indicating that CPCRNet has higher registration accuracy and better robustness compared to existing algorithms.

Problem statement: In facial biometrics, face features are used as the required human traits for automatic recognition. Feature extracted from face images are significant for face biometrics system performance. Approach: In this thesis, a framework of facial biometric was designed based on two subspace methods i.e., Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). First, PCA is used for dimension reduction, where original face images are projected into lower-dimensional face representations. Second, LDA was proposed to provide a solution of better discriminant. Both PCA and LDA features were presented to Euclidean distance measurement which is conveniently used as a benchmark. The algorithms were evaluated in face identification and verification using a standard face database-AT and T and a locally collected database-CBE. Each database consists of 400 images and 320 images respectively. Results: LDA-based methods outperform PCA for both face identification and verification. For face identification, PCA achieves accuracy of 91.9% (AT and T) and 76.7% (CBE) while LDA 94.2% (AT and T) and 83.1% (CBE). For face verification, PCA achieves Equal Error Rate (EER) of 1.15% (AT and T), 7.3% (CBE) while LDA 0.78% (AT and T) and 5.81% (CBE). Conclusion/Recommendations: This study had proved that, when given sufficient training samples, LDA is able to provide better discriminant ability in feature extraction for face biometrics.

Time-frequency methods for vibration-based gearbox faults detection have been considered the most efficient method. Among these methods, continuous wavelet transform (CWT) as one of the best time-frequency method has been used for both stationary and transitory signals. Some deficiencies of CWT are problem of overlapping and distortion ofsignals. In this condition, a large amount of redundant information exists so that it may cause false alarm or misinterpretation of the operator. In this paper a modified method called Exact Wavelet Analysis is used to minimize the effects of overlapping and distortion in case of gearbox faults. To implement exact wavelet analysis, Particle Swarm Optimization (PSO) algorithm has been used for this purpose. This method have been implemented for the acceleration signals from 2D acceleration sensor acquired by Advantech PCI-1710 card from a gearbox test setup in Amirkabir University of Technology. Gearbox has been considered in both healthy and chipped tooth gears conditions. Kernelized Support Vector Machine (SVM) with radial basis functions has used the extracted features from exact wavelet analysis for classification. The efficiency of this classifier is then evaluated with the other signals acquired from the setup test. The results show that in comparison of CWT, PSO Exact Wavelet Transform has better ability in feature extraction in price of more computational effort. In addition, PSO exact wavelet has better speed comparing to Genetic Algorithm (GA) exact wavelet in condition of equal population because of factoring mutation and crossover in PSO algorithm. SVM classifier with the extracted features in gearbox shows very good results and its ability has been proved.

A hybrid modeling algorithm based on partial least squares and neural network (BP algorithm) is proposed. First it extracts the feature from the original sample sets by partial least squares mehtod, and then the neural network regression using the extraction sets obtained is performed. Thus the hybrid modeling algorithm has the ability of feature extraction. The experiments results on the properties of engineering materials shows that the proposed hybrid algorithm can effectively modeling the properties of engineering materials with merits of dimensions reduction, elimination of noise and multiple correlations between independent variables.

Feature extraction is an important research topic in hyper-spectral image (HSI) classification. However, most of feature extraction methods only extract low-level features, which makes them not perform well in the applications of HSI. In this paper, we have proposed a non-negative matrix factorization (NMF) based deep feature extraction algorithm, namely deep NMF. Deep NMF tries to construct a deep feature representation by cascading multiple NMFs. Reconstruction residual of NMF is passed layer by layer to reduce information loss. Meanwhile, passing residuals between layers can construct a feature hierarchy from coarse to fine. Furthermore, activation functions are applied between adjacent layers to enhance the ability of non-linear feature extraction. Experimental results have also shown that our algorithm is computationally efficient and effective for HSI classification.

Changes on lakes and rivers are of great significance for the study of global climate change. Accurate segmentation of lakes and rivers is critical to the study of their changes. However, traditional water area segmentation methods almost all share the following deficiencies: high computational requirements, poor generalization performance, and low extraction accuracy. In recent years, semantic segmentation algorithms based on deep learning have been emerging. Addressing problems associated to a very large number of parameters, low accuracy, and network degradation during training process, this paper proposes a separable residual SegNet (SR-SegNet) to perform the water area segmentation using remote sensing images. On the one hand, without compromising the ability of feature extraction, the problem of network degradation is alleviated by adding modified residual blocks into the encoder, the number of parameters is limited by introducing depthwise separable convolutions, and the ability of feature extraction is improved by using dilated convolutions to expand the receptive field. On the other hand, SR-SegNet removes the convolution layers with relatively more convolution kernels in the encoding stage, and uses the cascading method to fuse the low-level and high-level features of the image. As a result, the whole network can obtain more spatial information. Experimental results show that the proposed method exhibits significant improvements over several traditional methods, including FCN, DeconvNet, and SegNet.

In response to the problems of complex background, multi-scale dangerous goods and severe stacking in X-ray security images, this paper proposes a high-accuracy dangerous goods detection algorithm for X-ray security images based on the improvement of YOLOv7. Firstly, by combining the coordinate attention mechanism, the downsampling structure of the backbone network is improved to enhance the model’s target feature localization ability. Secondly, a weighted bidirectional feature pyramid network is used as the feature fusion structure to achieve multi-scale feature weighted fusion and further simplify the network. Then, combined with dynamic snake convolution, a downsampling structure was designed to facilitate the extraction of features at different scales, providing richer feature representations. Finally, drawing inspiration from the idea of group convolution and combining it with Conv2Former, a feature extraction module called a multi-convolution transformer (MCT) was designed to enhance the network’s feature extraction ability by combining multi-scale information. The improved YOLOv7 in this article was tested on the public datasets SIXRay, CLCXray, and PIDray. The average detection accuracy (mAP) of the improved model was 96.3%, 79.3%, and 84.7%, respectively, which was 4.7%, 2.7%, and 3.1% higher than YOLOv7. This proves the effectiveness and universality of the method proposed in this article. Compared to the current mainstream X-ray image dangerous goods detection models, this model effectively reduces the false detection rate of dangerous goods in X-ray security inspection images and has achieved significant improvement in the detection of small and multi-scale targets, achieving higher accuracy in dangerous goods detection.

In the field of bearing defect detection, Aiming at the problem of low efficiency in manual inspection and prone to missed detections in scenarios with small target defects, and overlapping targets, an improved YOLOv5-based object detection method is proposed. Firstly, in terms of feature extraction, the C3 modules in the original backbone of YOLOv5 are replaced with the finer-grained Res2Block modules to improve the model’s feature extraction ability. Secondly, in terms of feature fusion, a Bidirectional Feature Pyramid Network (BiFPN) is added to the original neck of YOLOv5 to enhance the fusion ability of shallow graphic features and deep semantic features. Finally, the performance of the improved YOLOv5 algorithm is validated through ablation experiments and comparative experiments with other defect detection algorithms, including the Small_obj algorithm the existing method of adding a small target detection head for identifying small target defects. The experimental results demonstrate that the improved YOLOv5 algorithm exhibits high mAP and accuracy in bearing defect detection, enabling more precise identification the types of small target defects on bearings in complex scenarios with multiple coexisting defects and overlapping detection targets, thereby providing valuable reference for practical bearing defect detection.

To improve the feature extraction ability and shorten the learning time, fuzzy removing redundancy restricted Boltzmann machine (F3RBM) is developed. The features extracted by F3RBM with unsupervised learning are imported into support vector machine (SVM) to establish F3RBM-SVM model, which achieves fast and high-precision automatic classification of different samples. To expand the feature extraction capability of restricted Boltzmann machine (RBM), the deterministic parameters of control model are replaced by fuzzy numbers in view of the superiority of fuzzy idea and the redundancy removal mechanism is introduced. Comparing the feature similarity of hidden units with the threshold value, if the similarity is greater than the threshold value, they are considered to be redundant units with the same features. The redundant units are removed to achieve further dimension reduction. Finally, the learning speed, feature extraction ability, and classification accuracy of different models are compared in MINIST handwritten dataset, Fashion MNIST dataset, and Olivetti Face dataset. The experimental results show that the feature extraction capability of FRBM and F3RBM is better than that of RBM. When there are a large number of hidden units, the learning speed of F3RBM is obviously faster than that of FRBM. The features extracted from F3RBM are imported into the SVM to build F3RBM-SVM model, which improves the classification accuracy and learning speed than general classifier. When adding other noises, F3RBM-SVM has better robustness than other models.

Drawing the clothing plan is an essential part of the clothing industry. However, the irregular shape of clothing, strong deformability and sensitivity to light make the fast and accurate realization of clothing image retrieval a very challenging problem. The successful application of the Transformer in image recognition shows the application potential of the Transformer in the image field. This article proposes an efficient and improved clothing plan based on ResNet-50. Firstly, in the feature extraction section, the ResNet-50 network structure embedded in the Transformer module is used to improve the network's receptive field range and feature extraction ability. Secondly, dense jump connections are added to the ResNet-50 upsampling process, making full use of feature extraction information at each stage, further improving the quality of the generated image. The network consists of three steps: the sketch stage, which aims to predict the color distribution of clothing and obtain watercolor images without gradients and shadows. The second is the thinning stage, which refines the watercolor image into a clothing image with light and shadow effect; The third is the optimization stage, which combines the outputs of the first two stages to optimize the generation quality further. The experimental results show that the improved network's IS and first input delay (FID) scores are 4.592 and 1.506, respectively. High-quality clothing images can be generated only by inputting line drawings and a few color points. Compared with the existing methods, the image generated by this network has excellent advantages in realism and accuracy. This method can combine various feature information of images, improve retrieval accuracy, has strong robustness and practicability, and can provide a reference for the daily work of fashion designers.

Aiming at the problems of high labor cost, low detection efficiency, and insufficient detection accuracy of traditional pipe gallery disease detection methods, this paper proposes a pipe gallery disease segmentation model, PipeU-NetX, based on deep learning technology. By introducing the innovative down-sampling module MD-U, up-sampling module SC-U, and feature fusion module FFM, the model optimizes the feature extraction and fusion process, reduces the loss of feature information, and realizes the accurate segmentation of the pipe gallery disease image. In comparison with U-Net, FCN, and Deeplabv3+ models, PipeU-NetX achieved the best PA, MPA, FWIoU, and MIoU, which were 99.15%, 92.66%, 98.34%, and 87.63%, respectively. Compared with the benchmark model U-Net, the MIoU and MPA of the PipeU-NetX model increased by 4.64% and 3.92%, respectively, and the number of parameters decreased by 23.71%. The detection speed increased by 22.1%. The PipeU-NetX model proposed in this paper shows the powerful ability of multi-scale feature extraction and defect area adaptive recognition and provides an effective solution for the intelligent monitoring of the pipe gallery environment and accurate disease segmentation.