Target Detection In Images Research Articles

BM3D denoising-based multi-target detection method for complex background radar images

Abstract For the complex background of traditional radar image target detection methods is affected by multiple interferers, which leads to a low recognition rate of multi-target classification and a high false alarm rate. In this paper, the current hot research BM3D denoising method is introduced into radar image target detection. Firstly, the detection process of the BM3D algorithm is constructed, and the image pixels are reasonably clustered by denoising so that the image blocks in homogeneous regions have a high matching degree and those in non-homogeneous regions have a low matching degree. Secondly, the clustering recognition feature is used to limit the search range of similar blocks to homogeneous regions, to improve the denoising efficiency of the algorithm. Finally, the simulation results of this paper show that compared with previous algorithms, the BM3D detection method has significantly improved in both the subjective visual effect map and objective numerical indexes. For example, the image denoising quality is improved by about 0.716dB on average and up to 1.031dB, the image denoising time is shortened by about 19.962s on average, and the algorithm runs 1.211 times faster than the original algorithm. It is proved that the algorithm can improve denoising efficiency and reduce computational complexity while enhancing detailed information.

Visualization of green building landscape space environment design based on image processing and artificial intelligence algorithm

The development direction of future architectural design should be in line with environmental protection, green and energy conservation, so as to meet people’s sustainable development needs. The promotion of green building technology needs to involve multi-level and multi-disciplinary knowledge and social factors. Through a large number of green building projects, we can know that its ecological landscape not only has artistic value, but also can reflect obvious ecological value. An important concept in the concept of sustainable development is the green building landscape. Based on this background, this paper introduces artificial intelligence algorithm and designs a visual system around the concept of landscape space environment design. This system can be operated in two different forms, in offline application, image training is mainly carried out on the server side; in the online stage, the collected images are uploaded to the server using the Android platform, and the server side performs a series of operations such as image target detection, and then uses the Android mobile terminal to display the virtual reality fusion scene. The system test shows that the green building environment visualization system can not only make full use of the idle time of the cable drainage reader, but also will not affect the search operation in the next stage. The average delay is reduced by 16%, which proves the practicability of the software. In this paper, artificial intelligence algorithm is introduced into the field of green landscape space environment design, and an effective visualization system is constructed.

Rapeseed Storage Quality Detection Using Hyperspectral Image Technology - An Application for Future Smart Cities

Abstract At present, the application of hyperspectral image technology in image target detection is lacking black-and-white correction, and the average spectral reflectance cannot be calculated, which leads to large error in image feature detection and classification. In this study, hyperspectral image technology was applied to the detection of rapeseed storage quality, and germination detection was completed during the storage of rapeseed. The black-and-white board correction to the hyperspectral data was completed and the spectral characteristic curve of the rapeseed sample hyperspectral image was obtained. The average spectral reflectance is calculated, the threshold of hyperspectral image is estimated, and the correlation technique is used to denoise the hyperspectral image. Based on this, the edge feature of the rapeseed hyperspectral image is recognized, and the feature classification of the hyperspectral rapeseed image is realized by combining the gray co-occurrence matrix. The experimental results show that the proposed method can detect the germination of rapeseed with high precision under the application of hyperspectral image technology. This study provides a reliable basis for the application of hyperspectral image technology.

Deep Classification with Linearity-Enhanced Logits to Softmax Function.

Recently, there has been a rapid increase in deep classification tasks, such as image recognition and target detection. As one of the most crucial components in Convolutional Neural Network (CNN) architectures, softmax arguably encourages CNN to achieve better performance in image recognition. Under this scheme, we present a conceptually intuitive learning objection function: Orthogonal-Softmax. The primary property of the loss function is to use a linear approximation model that is designed by Gram-Schmidt orthogonalization. Firstly, compared with the traditional softmax and Taylor-Softmax, Orthogonal-Softmax has a stronger relationship through orthogonal polynomials expansion. Secondly, a new loss function is advanced to acquire highly discriminative features for classification tasks. At last, we present a linear softmax loss to further promote the intra-class compactness and inter-class discrepancy simultaneously. The results of the widespread experimental discussion on four benchmark datasets manifest the validity of the presented method. Besides, we want to explore the non-ground truth samples in the future.

An Improved YOLO Model for UAV Fuzzy Small Target Image Detection

High-altitude UAV photography presents several challenges, including blurry images, low image resolution, and small targets, which can cause low detection performance of existing object detection algorithms. Therefore, this study proposes an improved small-object detection algorithm based on the YOLOv5s computer vision model. First, the original convolution in the network framework was replaced with the SPD-Convolution module to eliminate the impact of pooling operations on feature information and to enhance the model’s capability to extract features from low-resolution and small targets. Second, a coordinate attention mechanism was added after the convolution operation to improve model detection accuracy with small targets under image blurring. Third, the nearest-neighbor interpolation in the original network upsampling was replaced with transposed convolution to increase the receptive field range of the neck and reduce detail loss. Finally, the CIoU loss function was replaced with the Alpha-IoU loss function to solve the problem of the slow convergence of gradients during training on small target images. Using the images of Artemisia salina, taken in Hunshandake sandy land in China, as a dataset, the experimental results demonstrated that the proposed algorithm provides significantly improved results (average precision = 80.17%, accuracy = 73.45% and recall rate = 76.97%, i.e., improvements by 14.96%, 6.24%, and 7.21%, respectively, compared with the original model) and also outperforms other detection algorithms. The detection of small objects and blurry images has been significantly improved.

Target Detection for Motion Images Using the Improved YOLO Algorithm

The images of motion states are time-varying, and when actually detecting their internal motion targets, the formed detection frames overlap, resulting in small confidence values for the detection frames and low accuracy of the detection results. To address this problem, the authors propose a target detection for motion image using the improved YOLO algorithm. First, the YOLO algorithm is improved using deformable convolution; the edge weights of the front and back views within the image are collated, and the motion image is segmented using the improved YOLO algorithm. Second, the structure formed by the initial convolution is used as the initial detection frame structure, the parallel cross-ratio value is set, the overlap generated by the detection frame is controlled, the parameters of the detection frame compression processing are output, the threshold trigger value relationship is constructed, and finally, the detection of the motion image target is realized. The results show that the target false detection rate of the proposed method is only about 15%. The detection a priori frame height value is 80 pixels, and the average detection time consumed is 6.8ms, which proves that the proposed algorithm can be widely used in motion image target detection to improve the detection level.

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.

Grid-guided localization network based on the spatial attention mechanism for synthetic aperture radar ship detection

Synthetic aperture radar (SAR) has become an important part of ship detection due to its all-day all-weather characteristics. Ship monitoring is important for coastal traffic control and territorial safety. Ship detection requires high accuracy, but the speed of ship detection is also an important parameter in making rapid decisions, such as those in maritime rescue and military strategy. However, fast networks tend to be less accurate. To address this problem, we propose an efficient localization method based on a spatial attention mechanism. This mechanism uses grid-guided localization instead of a frame regression mechanism, which makes the network faster but also lowers the accuracy to a level insufficient for direct application to SAR image target detection. A tested and selected spatial attention mechanism improves the detection accuracy while guaranteeing its speed. It is shown that four factors affect the design of the spatial attention mechanism, namely query and key content, query content and relative position, key content, and relative position. These factors are ordered, and the group with the best precision is added to the proposed network. The proposed network can achieve a good detection effect for sparsely distributed targets in SAR images, and the proposed algorithm can achieve AP50 ∶ 75, AP50, AP75, and APS values of 71.3%, 96.9%, 86.6%, and 72.1%, respectively, on the SAR ship detection dataset.

Research on computer vision technology based on BP-LSTM hybrid network

Abstract The computer vision direction in the field of artificial intelligence analyses the latest progress of computer vision technology from visual perception and visual generation, including but not limited to image recognition, target detection and image segmentation. First of all, for computer vision technology, this paper introduces the detailed application of image recognition technology, object detection technology and image segmentation technology. Then, we build a BP neural network combined with a deep LSTM neural network, use the BP network algorithm to select the input variables to reduce the dimension and complexity of the model, and use the selected variables as the input of the deep LSTM network. At the same time, deep LSTM is used to perform high-dimensional deep memory learning features on the selected variables. Finally, the model is separately experimented in computer vision. The experimental results show that the present model and other single models can be selected by BP neural network variables in computer vision applications, which can effectively reduce the complexity of the model and improve the generalisation ability of the model, so that it can be used in computer vision research.

BoT2L-Net: Appearance-Based Gaze Estimation Using Bottleneck Transformer Block and Two Identical Losses in Unconstrained Environments

As a nonverbal cue, gaze plays a critical role in communication, expressing emotions and reflecting mental activity. It has widespread applications in various fields. Recently, the appearance-based gaze estimation method, which utilizes CNN (convolutional neural networks), has rapidly improved the accuracy and robustness of gaze estimation algorithms. Due to their insufficient ability to capture global relationships, the present accuracy of gaze estimation methods in unconstrained environments, has the potential for improvement. To address this challenge, the focus of this paper is to enhance the accuracy of gaze estimation, which is typically measured by mean angular error. In light of Transformer’s breakthrough in image classification and target detection tasks, and the need for an efficient network, the Transformer-enhanced-CNN method is a suitable choice. This paper proposed a novel model for 3D gaze estimation in unconstrained environments, based on the Bottleneck Transformer block and multi-loss methods. Our designed network (BoT2L-Net), incorporates self-attention through the BoT block, utilizing two identical loss functions to predict the two gaze angles. Additionally, the back-propagation network was combined with classification and regression losses, to improve the network’s accuracy and robustness. Our model was evaluated on two commonly used gaze datasets: Gaze360 and MPIIGaze, achieving mean angular errors of 11.53° and 9.59° for front 180° and front-facing gaze angles, respectively, on the Gaze360 testing set, and a mean angular error of 3.97° on the MPIIGaze testing set, outperforming the CNN-based gaze estimation method. The BoT2L-Net model proposed in this paper performs well on two publicly available datasets, demonstrating the effectiveness of our approach.

A dataset of desert plant images for deep learning recognition in Xinjiang in 2020–2021

Automatic recognition of desert plant types by machine vision can support the research on wind prevention and sand fixation, ecosystem value assessment, vegetation restoration and reconstruction, and reduce the dependence on plant expert identification. At present, the research on the machine discrimination model of desert plants mainly relies on the standardized high-quality plant specimen images, lacking the desert plant images obtained under complex natural conditions. This dataset provides typical desert plant images of Xinjiang that can be used for the model training of deep learning image classification, including 15,550 digital camera images of desert plants in Xinjiang obtained under different seasons, natural backgrounds and lighting conditions, and covering 19 typical desert plant types. Suaeda salsa has the smallest number of images and Artemisia desertorum has the biggest, 465 and 1,240 respectively, with a median of 800, which has met the training needs of mainstream deep learning model. This dataset can provide basic data for desert plant image segmentation, target detection and automatic recognition.

MIoP-NMS: Perfecting crops target detection and counting in dense occlusion from high-resolution UAV imagery

When performing dense crops target detection and counting tasks in high-resolution images of large scene, several algorithms such as the standard Non-Maximum Suppression (NMS) and its improved CIoU-NMS and soft-NMS are commonly used to improve the de-duplication performance of the prediction frame in objects detection, which has a high mean average precision (mAP) but the error boundaries may be large. In this research, an improved NMS-based max intersection over portion (MIoP-NMS) algorithm is proposed to address this problem and implemented in the YOLOv4 network framework for single-stage target detection. Using banana tree target detection statistics with different sparsity levels as an experimental case, the mAP obtained by applying the method in the paper reaches about 85%, with a mean counting error of about 1.3%. In comparison with several commonly NMS methods, their mAP values are comparable. However, compared to the 10%∼25% mean count error when using these commonly used methods, the mean count error of the method proposed in the paper is reduced by about 8.7%∼23.7%. The method in the paper estimates the number of banana trees in dense occluded banana forests with about 98.7% accuracy, which is simple and effective, and has significant advantages for accurate detection and counting of dense crops targets in high-resolution images in large scenes.

FPGA Implementation of a Deep Learning Acceleration Core Architecture for Image Target Detection

Due to the flexibility and ease of deployment of Field Programmable Gate Arrays (FPGA), more and more studies have been conducted on developing and optimizing target detection algorithms based on Convolutional Neural Networks (CNN) models using FPGAs. Still, these studies focus on improving the performance of the core algorithm and optimizing hardware structure, with few studies focusing on the unified architecture design and corresponding optimization techniques for the algorithm model, resulting in inefficient overall model performance. The essential reason is that these studies do not address arithmetic power, speed, and resource consistency. In order to solve this problem, we propose a deep learning acceleration core architecture based on FPGAs, which is designed for target detection algorithms with CNN models, using multi-channel parallelization of CNN network models to improve the arithmetic power, using scheduling tasks and intensive computation pipelining to meet the algorithm’s data bandwidth requirements and unifying the speed and area of the orchestrated computation matrix to save hardware resources. The proposed framework achieves 14 Frames Per Second (FPS) inference performance of the TinyYolo model at 5 Giga Operations Per Second (GOPS) with 30% higher running clock frequency, 2–4 times higher arithmetic power, and 28% higher Digital Signal Processing (DSP) resource utilization efficiency using less than 25% of FPGA resource usage.

Using deep learning in an embedded system for real-time target detection based on images from an unmanned aerial vehicle: vehicle detection as a case study

ABSTRACT For a majority of remote sensing applications of unmanned aerial vehicles (UAVs), the data need to be downloaded to ground devices for processing, but this procedure cannot satisfy the demands of real-time target detection. Our objective in this study is to develop a real-time system based on an embedded technology for image acquisition, target detection, the transmission and display of the results, and user interaction while providing support for the interactions between multiple UAVs and users. This work is divided into three parts: (1) We design the technical procedure and the framework for the implementation of a real-time target detection system according to application requirements. (2) We develop an efficient and reliable data transmission module to realize real-time cross-platform communication between airborne embedded devices and ground-side servers. (3) We optimize the YOLOv4 algorithm by using the K-Means algorithm and TensorRT inference to improve the accuracy and speed of the NVIDIA Jetson TX2. In experiments involving static detection, it had an overall confidence of 89.6% and a rate of missed detection of 3.8%; in experiments involving dynamic detection, it had an overall confidence and a rate of missed detection of 88.2% and 4.6%, respectively.

Moving Targets Artifacts Removal in Multiaspect SAR Imagery Based on Logarithm Background Subtraction

Multiaspect SAR has the capability of providing a high-resolution image due to its long synthetic aperture feature. However, a moving target can generate long and complex signatures in a multiaspect SAR image, which may hamper the applications like image interpretation and target detection. In this article, two methods are proposed to remove the moving target signature in a single-channel multiaspect SAR image. The two methods are all based on logarithm background subtraction. The first one is a fast scheme with a cost of reduced resolution. While the second one focuses on preserving the high resolution, it takes more time than the previous one. The first method utilizes the fact of target signal position changes in subaperture image sequence to obtain the static background. The second method combines the detection results to exclude the moving target signal in each complex-valued subaperture image, then obtaining the high resolution of static background by coherent summation. The methods are validated by synthetic and real airborne SAR data.

Infrared object detection via patch‐tensor model and image denoising based on weighted truncated Schatten‐p norm minimization

Infrared object detection via patch‐tensor model and image denoising based on weighted truncated Schatten‐<i>p</i> norm minimization

Sonar image target detection based on multi-region optimal selection strategy

To overcome the adverse effects of noise and shadow regions on target detection in side-scan sonar images, more precisely, it is difficult to accurately detect targets, a target detection technology based on a multi-region optimal selection strategy of spectral clustering combined with the entropy weight method is proposed in this study. First, the cluster numbers for spectral clustering are set in advance based on prior knowledge, and the pixels of the sonar image are clustered into several different regions. Second, the invariable features of translation, rotation and scaling up that each region is extracted and used to construct the feature criterion matrix for the multiple regions. Last, the entropy weight method is used to calculate the weights of each feature and the comprehensive weighted score of each region for this feature criterion matrix to obtain the final target region. Experimental results show that the proposed method can effectively overcome the adverse effects of noise and shadow regions in side-scan sonar images, but also achieve the selection of optimal target region among multiple regions after image clustering, thus verifying the feasibility and effectiveness of the proposed method in this study.

A Reverberation Noise Suppression Method of Sonar Image Based on Shearlet Transform

A sonar image affected by reverberation noise would generate serious speckle noise in the image, which causes great trouble for the later sonar image target segmentation and detection. Therefore, noise suppression is a crucial preprocessing technique for sonar image applications. To suppress the reverberation noise effectively in the sonar image, a reverberation noise suppression method of sonar image based on shearlet transform is proposed. First, to effectively apply the noise suppression method to sonar images, a method of combining the transform from the reverberation noise model to the Gaussian additive noise model with the shearlet transform is presented. Second, in the process of the shearlet transform, constructing a shear wave filter based on the Meyer window function, which has an excellent performance in both time and frequency domains, is suggested. Third, a noise variance estimation based on weak texture blocks would be made to avoid the effect of gray overflow in sonar images. Then, the estimated variance is combined with the root mean square coefficient obtained from the shearlet transform of the simulated noise. For this reason, obtaining the adaptive filtering thresholds of each scale and each direction is essential. With this process, a better reverberation suppression effect could be achieved. Simulation and experimental results demonstrate that the method proposed in this article can suppress reverberation noise more effectively, and a better edge-preserving effect could be obtained. Compared with other comparison algorithms, the algorithm proposed in this article could achieve the best results on the peak signal-to-noise ratio (PSNR) and SSMI.

UAV aerial image target detection based on BLUR-YOLO

ABSTRACT With the development of unmanned aerial vehicle (UAV) remote sensing technology, target detection based on UAV images has increasingly become a hot spot for research. Aiming at the problems of many small target instances, complex backgrounds and difficult feature extraction in UAV images, we propose a UAV aerial image target detection algorithm called BLUR-YOLO. First, the h-swish activation function is used in the backbone network and the neck network to increase the expressiveness of the model. Second, an attention mechanism (CoordAttention) is added to the bottleneck layer of the backbone network, thereby increasing the weight of valid information and suppressing background noise interference. Finally, by removing redundant nodes of the path aggregation network (PANet), adding additional connections, and using BlurPool instead of the downsampling method, a feature pyramid network (Blur-PANet) is proposed to effectively fuse multilayer features. Experimental results on the VisDrone public dataset show that the proposed drone image object detection model (BLUR-YOLO) is 1.2% better than the YOLOv4 algorithm, which proves the effectiveness of the method.

Infrared Image Target Detection of Substation Electrical Equipment Using an Improved Faster R-CNN

Infrared camera can be used to monitor the condition of substation electrical equipment. Fast and accurate target detection algorithm is the key for infrared intelligent on-line routing inspection. However, the performance of traditional detection algorithm is poor due to the complex background of substation. To solve this problem, this paper proposes a target detection model based on the improved faster region-based convolutional neural networks (Faster R-CNN), which can be used for automatic detection of five kinds of electrical equipment in substations. The feature extraction network of this model is improved based on VGG16 by abandoning some high-level convolutions to accelerate the training and testing speed. Meanwhile, the 1:3 and 3:1 aspect ratio of anchor are added to improve the detection accuracy of elongated equipment. Experiments are performed on an infrared image dataset of substation to detect five types of equipment. The robustness tests of our model are carried out, too. The results show that our model performs well in detection accuracy and speed, achieving a mean detected accuracy of 95.32% and running at 11 frame/second. In addition, our model is robust to noise and lightness, which is suitable to other substations. Comparing to other models, our model has the highest mAP@0.5 of 92.78%.