Combined Loss Function Research Articles

A physics knowledge-based surrogate model framework for time-dependent slope deformation: Considering water effect and sliding states

The surrogate model serves as an efficient simulation tool during the slope parameter inversion process. However, the creep constitutive model integrated with dynamic damage evolution poses challenges in development of the required surrogate model. In this study, a novel physics knowledge-based surrogate model framework is proposed. In this framework, a Transformer module is employed to capture strain-driven softening-hardening physical mechanisms. Positional encoding and self-attention are utilized to transform the constitutive parameters associated with shear strain, which are not directly time-related, into intermediate latent features for physical loss calculation. Next, a multi-layer stacked GRU (gated recurrent unit) network is built to provide input interfaces for time-dependent intermediate latent features, hydraulic boundary conditions, and water-rock interaction degradation equations, with static parameters introduced via external fully-connected layers. Finally, a combined loss function is constructed to facilitate the collaborative training of physical and data loss, introducing time-dependent weight adjustments to focus the surrogate model on accurate deformation predictions during critical phases. Based on the deformation of a reservoir bank landslide triggered by impoundment and subsequent restabilization, an elasto-viscoplastic constitutive model that considers water effect and sliding state dependencies is developed to validate the proposed surrogate model framework. The results indicate that the framework exhibits good performance in capturing physical mechanisms and predicting creep behavior, reducing errors by about 30 times compared to baseline models such as GRU and LSTM (long short-term memory), meeting the precision requirements for parameter inversion. Ablation experiments also confirmed the effectiveness of the framework. This framework can also serve as a reference for constructing other creep surrogate models that involve non-time-related across dimensions.

ADD-YOLO: An algorithm for detecting animals in outdoor environments based on unmanned aerial imagery

The breeding management of extensive livestock and scientific research surveys of animals in outdoor environments often require the utilization of UAVs due to their ability to efficiently cover large areas at a cost-effective rate. However, identifying small animal targets in aerial imagery from high-altitudes remains a significant challenge. This paper introduces an enhanced algorithm based on YOLOv8n, specifically designed for aerial animal detection. Firstly, we add a P2 small target detection layer on top of the original baseline model, while removing the P5 large target detection layer and 32x downsampling to enhance the detection of small animal targets and reduce the number of model parameters. Secondly, The improved N-SPPCSPC module replaces the spatial pyramid pooling structure in the baseline model to enhance the extraction capability for small targets. Thirdly, an improved DWRC2f module is adopted to enhance the extraction of multi-scale contextual information. Fourthly, the SEAM module is incorporated before the detection head to enhance the detection of occluded and overlapping animals. Finally, a combined NWD Loss function is implemented to address the scale sensitivity of IoU Loss, thereby improving the accuracy of small target detection. Compared to the baseline model, the improved model achieved an increase of 7.1% and 4.9% in mAP50 values and an increase of 4.0% and 1.3% in mAP50-95 values, respectively, across two datasets, while significantly reducing the number of parameters. Further comparisons with other single-stage object detection models demonstrate a better robustness of our model. Additionally, after quantization, when testing the inference speed of ADD-YOLO on performance-constrained edge devices, it was 1.72 times and 1.81 times faster than the baseline model. Therefore, this model provides a new and efficient monitoring tool for extensive pastoral management and wildlife surveys.

Phase unwrapping in digital holography based on SRDU-net

In digital holographic measurement, the hologram phase is extracted using an inverse tangent function, resulting in a wrapped phase that is constrained to be (-π,π]. However, the phase contains the height information of the object, which is crucial for accurate measurement of the object contour. Therefore, a digital holographic phase unwrapping method based on SRDU-Net (Separable-Residual-Dense-Inverted U-Net) is proposed in this paper. The SRDU-Net network is constructed by introducing depth-separable convolution, inverted residuals and dense blocks with U-Net as the network framework to achieve high-precision hologram phase recovery. This network adopts depth-separable convolution instead of traditional convolution, combines the inverse residual connection to construct a lightweight convolution structure, and defines dense blocks with this structure to form a lightweight grouped deep convolution network. Meanwhile, the Leaky ReLU activation function is used to introduce the learning rate mechanism to optimize the network parameters with Huber and MSE (mean square error) as the combined loss function. The simulated phase dataset is used to train the network, the trained network model is tested for speckle noise immunity, and phase unwrapping experiments are performed on the collected holograms of the test samples. The results show that SRDU-Net improves the SSIM by 0.1% and reduces the RMSE by 86% over Res-UNet (Residual-UNet). Therefore, the proposed method can realize high-precision recovery of digital holographic wrapped phases, and has a good robustness to phase unwrapping of holograms containing a high degree of speckle noise.

3D Face Reconstruction Based on ResNet Feature Extraction and CBAM

In view of the scarcity, high cost and lack of diversity of three-dimensional (3D) face datasets, this paper designs an end-to-end self-supervised learning 3D face reconstruction network, which uses single 2D face image as input. The model bypasses the 3D face datasets and only uses the 2D face datasets for training to achieve high-precision 3D face reconstruction without any 3D face prior. First, the improved ResNet50 feature extraction module is introduced to extract and characterize the input image by deep convolutional network. Then, a lightweight convolutional block attention module is added to the face prediction subnetwork. On the one hand, channel attention extracts different information included in the image, and on the other hand spatial attention finds the location of the information. So, the serialized attention operation could accurately find the features required for different parameter predictions, further improving face reconstruction parameters’ prediction accuracy. Finally, training, ablation and comparison experiments were conducted on CelebFaces Attributes, basel face model and Photoface datasets, and the combined loss function of pixel loss and perception loss was selected. The pixel loss function was calculated at the pixel microscopic level, and the perception loss function was calculated at the image macroscopic convolution level. The combination of the two could complement each other. Compared with the historical optimal results of the same network structure, the scale-invariant depth error and mean angle deviation of the proposed algorithm are improved by 5.2% and 8.2%, respectively. Experimental results strongly prove the effectiveness of the algorithm.

Mobile User Traffic Generation Via Multi-Scale Hierarchical GAN

Mobile user traffic facilitates diverse applications, including network planning and optimization, whereas large-scale mobile user traffic is hardly available due to privacy concerns. One alternative solution is to generate mobile user traffic data for downstream applications. However, existing generation models cannot simulate the multi-scale temporal dynamics in mobile user traffic on individual and aggregate levels. In this work, we propose a multi-scale hierarchical generative adversarial network (MSH-GAN) containing multiple generators and a multi-class discriminator. Specifically, the mobile traffic usage behavior exhibits a mixture of multiple behavior patterns, which are called micro-scale behavior patterns and are modeled by different pattern generators in our model. Moreover, the traffic usage behavior of different users exhibits strong clustering characteristics, with the co-existence of users with similar and different traffic usage behaviors. Thus, we model each cluster of users as a class in the discriminator’s output, referred to as macro-scale user clusters. Then, the gap between micro-scale behavior patterns and macro-scale user clusters is bridged by introducing the switch mode generators, which describe the traffic usage behavior in switching between different patterns. All users share the pattern generators. In contrast, the switch mode generators are only shared by a specific cluster of users, which models the multi-scale hierarchical structure of the traffic usage behavior of massive users. Finally, we urge MSH-GAN to learn the multi-scale temporal dynamics via a combined loss function, including adversarial loss, clustering loss, aggregated loss, and regularity terms. Extensive experiment results demonstrate that MSH-GAN outperforms state-of-art baselines by at least 118.17% in critical data fidelity and usability metrics. Moreover, observations show that MSH-GAN can simulate traffic patterns and pattern switch behaviors.

An Early Warning Model for Turbine Intermediate-Stage Flux Failure Based on an Improved HEOA Algorithm Optimizing DMSE-GRU Model

As renewable energy sources such as wind and photovoltaics continue to enter the grid, their intermittency and instability leads to an increasing demand for peaking and frequency regulation. An efficient dynamic monitoring method is necessary to improve the safety level of intelligent operation and maintenance of power stations. To overcome the insufficient detection accuracy and poor adaptability of traditional methods, a novel fault early warning method with careful consideration of dynamic characteristics and model optimization is proposed. A combined loss function is proposed based on the dynamic time warping and the mean square error from the perspective of both shape similarity and time similarity. A prediction model of steam turbine intermediate-stage extraction temperature based on the gate recurrent unit is then proposed, and the change in prediction residuals is utilized as a fault warning criterion. In order to further improve the diagnostic accuracy, a human evolutionary optimization algorithm with lens opposition-based learning is proposed for model parameter adaptive optimization. Experiments on real-world normal and faulty operational data demonstrate that the proposed method can improve the detection accuracy by an average of 1.31% and 1.03% compared to the long short-term memory network, convolutional neural network, back propagation network, extreme learning machines, gradient boosting decision tree, and LightGBM models.

Study on a Landslide Segmentation Algorithm Based on Improved High-Resolution Networks

Landslides are a kind of geological hazard with great destructive potential. When a landslide event occurs, a reliable landslide segmentation method is important for assessing the extent of the disaster and preventing secondary disasters. Although deep learning methods have been applied to improve the efficiency of landslide segmentation, there are still some problems that need to be solved, such as the poor segmentation due to the similarity between old landslide areas and the background features and missed detections of small-scale landslides. To tackle these challenges, a proposed high-resolution semantic segmentation algorithm for landslide scenes enhances the accuracy of landslide segmentation and addresses the challenge of missed detections in small-scale landslides. The network is based on the high-resolution network (HR-Net), which effectively integrates the efficient channel attention mechanism (efficient channel attention, ECA) into the network to enhance the representation quality of the feature maps. Moreover, the primary backbone of the high-resolution network is further enhanced to extract more profound semantic information. To improve the network’s ability to perceive small-scale landslides, atrous spatial pyramid pooling (ASPP) with ECA modules is introduced. Furthermore, to address the issues arising from inadequate training and reduced accuracy due to the unequal distribution of positive and negative samples, the network employs a combined loss function. This combined loss function effectively supervises the training of the network. Finally, the paper enhances the Loess Plateau landslide dataset using a fractional-order-based image enhancement approach and conducts experimental comparisons on this enriched dataset to evaluate the enhanced network’s performance. The experimental findings show that the proposed methodology achieves higher accuracy in segmentation performance compared to other networks.

Text-independent voiceprint recognition via compact embedding of dilated deep convolutional neural networks

In order to process speech, most state-of-the-art experimental methods employ convolutional neural networks (CNNs), which operate on a continuous, 1-dimensional (1-D) time stream. In an audio signal, the mel-spectrogram facilitates the representation of attributes of the utterances' in the frequency domain (which corresponds to the speech spectrum). Moreover, for a time-series speaker signal, CNNs are superior to machine or transfer learning models in capturing characteristics from long-form talks. This paper introduces a jump-connected 1-D CNN that employs a combined loss function for speaker recognition. The suggested model uses a 1-D convolutional layer combined with jump connections to extract speaker-specific characteristics; this reduces time-based and frequency-based variability for faster computing. A combined softmax loss, stable L2-norm, and smooth L1-norm loss function guide the proposed compact convolutional neural networks (CCNN) to identify the correct spokesman with improved efficacy. We evaluated the proposed framework using various standard and real-time audio datasets. The experimental findings demonstrate that the proposed CCNN outperforms existing approaches by reducing the equal error rate by 9.02 %. Also, our recommended voiceprint identification model achieves an impressive average speaker recognition rate of 98.76 %. Simultaneously, the reliability of the 1-D CCNN is tested under various conditions. Other fields of study, like language modelling, could employ this approach after some fine-tuning.Relevance of the work: Speaker recognition is an area of interest in which machine learning (ML) and deep learning (DL) schemes, when combined, have the potential to make history in the areas of forensic sciences, automation, and authentication. Using a modest CNN can enhance the identification and verification process by ignoring many issues such as false positives, background noise, and so on. Expanding this process would facilitate raga identification and disease treatment therapies.

Complex profile metrology via physical symmetry enhanced small angle x-ray scattering

Small angle x-ray scattering (SAXS) stands out as a promising solution in semiconductor metrology. The critical issue of SAXS metrology is to solve the SAXS inverse problem. With the increasing complexity of semiconductor devices, traditional strategies will face problems such as long iteration time and multiple solutions. To address these challenges, we develop a physical symmetry enhanced method to speed up the solution of the SAXS inverse problem for complex nanostructures. We incorporate the physical symmetry into a deep learning model, and a combined loss function is proposed to determine the correct structure in each step of training, which can continuously correct errors and make the model converge faster. The results show that the proposed method achieves high accuracy in determining the critical structural parameters of the complex profile gratings. Compared to traditional strategies, our method performs better in accuracy and does not require time-consuming iterations during reconstruction. The physical symmetry enhanced method provides a feasible way for achieving real-time reconstruction of complex profile nanostructures and is expected to promote the development of SAXS metrology.

An Efficient Semantic Segmentation Method for Remote-Sensing Imagery Using Improved Coordinate Attention

Semantic segmentation stands as a prominent domain within remote sensing that is currently garnering significant attention. This paper introduces a pioneering semantic segmentation model based on TransUNet architecture with improved coordinate attention for remote-sensing imagery. It is composed of an encoding stage and a decoding stage. Notably, an enhanced and improved coordinate attention module is employed by integrating two pooling methods to generate weights. Subsequently, the feature map undergoes reweighting to accentuate foreground information and suppress background information. To address the issue of time complexity, this paper introduces an improvement to the transformer model by sparsifying the attention matrix. This reduces the computing expense of calculating attention, making the model more efficient. Additionally, the paper uses a combined loss function that is designed to enhance the training performance of the model. The experimental results conducted on three public datasets manifest the efficiency of the proposed method. The results indicate that it excels in delivering outstanding performance for semantic segmentation tasks pertaining to remote-sensing images.

BDHE-Net: A Novel Building Damage Heterogeneity Enhancement Network for Accurate and Efficient Post-Earthquake Assessment Using Aerial and Remote Sensing Data

Accurate and efficient post-earthquake building damage assessment methods enable key building damage information to be obtained more quickly after an earthquake, providing strong support for rescue and reconstruction efforts. Although many methods have been proposed, most have limited effect on accurately extracting severely damaged and collapsed buildings, and they cannot meet the needs of emergency response and rescue operations. Therefore, in this paper, we develop a novel building damage heterogeneity enhancement network for pixel-level building damage classification of post-earthquake unmanned aerial vehicle (UAV) and remote sensing data. The proposed BDHE-Net includes the following three modules: a data augmentation module (DAM), a building damage attention module (BDAM), and a multilevel feature adaptive fusion module (MFAF), which are used to alleviate the weight deviation of intact and slightly damaged categories during model training, pay attention to the heterogeneous characteristics of damaged buildings, and enhance the extraction of house integrity contour information at different resolutions of the image. In addition, a combined loss function is used to focus more attention on the small number of severely damaged and collapsed classes. The proposed model was tested on remote sensing and UAV images acquired from the Afghanistan and Baoxing earthquakes, and the combined loss function and the role of the three modules were studied. The results show that compared with the state-of-the-art methods, the proposed BDHE-Net achieves the best results, with an F1 score improvement of 6.19–8.22%. By integrating the DBA, BDAM, and MFAF modules and combining the loss functions, the model’s classification accuracy for severely damaged and collapsed categories can be improved.

Cluster-Based Pairwise Contrastive Loss for Noise-Robust Speech Recognition.

This paper addresses a joint training approach applied to a pipeline comprising speech enhancement (SE) and automatic speech recognition (ASR) models, where an acoustic tokenizer is included in the pipeline to leverage the linguistic information from the ASR model to the SE model. The acoustic tokenizer takes the outputs of the ASR encoder and provides a pseudo-label through K-means clustering. To transfer the linguistic information, represented by pseudo-labels, from the acoustic tokenizer to the SE model, a cluster-based pairwise contrastive (CBPC) loss function is proposed, which is a self-supervised contrastive loss function, and combined with an information noise contrastive estimation (infoNCE) loss function. This combined loss function prevents the SE model from overfitting to outlier samples and represents the pronunciation variability in samples with the same pseudo-label. The effectiveness of the proposed CBPC loss function is evaluated on a noisy LibriSpeech dataset by measuring both the speech quality scores and the word error rate (WER). The experimental results reveal that the proposed joint training approach using the described CBPC loss function achieves a lower WER than the conventional joint training approaches. In addition, it is demonstrated that the speech quality scores of the SE model trained using the proposed training approach are higher than those of the standalone-SE model and SE models trained using conventional joint training approaches. An ablation study is also conducted to investigate the effects of different combinations of loss functions on the speech quality scores and WER. Here, it is revealed that the proposed CBPC loss function combined with infoNCE contributes to a reduced WER and an increase in most of the speech quality scores.

Shallow Inception Domain Adaptation Network for EEG-Based Motor Imagery Classification

Electroencephalography (EEG) data across multiple individuals have a high variance. Directly using the data to train a deep learning model usually degrades the performance. To address this issue, we propose a shallow Inception domain adaptation framework to extract informative deep features from data of multiple subjects for accurate motor imagery (MI) recognition. To our best knowledge, the Inception architecture in deep learning is combined with domain adaptation (DA) scheme for the first time for the MI classification task. The approach contains two compact Inception blocks that decode temporal features in different scales. In addition, we jointly optimize a novel combined loss function to reduce both marginal and class conditional discrepancies caused by the multi-modal structure of EEG signals. The DA-based loss enables Inception blocks to take full advantage of their learning abilities to capture discriminative patterns of MI data from multiple subjects instead of relying on the target user only. To demonstrate the effectiveness of our approach, we conduct substantial experiments on two wellknown datasets, BCI competition IV-2a and IV-2b. Results show that our model achieves better performance than state-of-theart strategies. The proposed model is able to extract informative features from high-variant EEG data collected from different individuals and achieves accurate MI classifications.

Hyperspectral image dynamic range reconstruction using deep neural network-based denoising methods

Hyperspectral (HS) measurement is among the most useful tools in agriculture for early disease detection. However, the cost of HS cameras that can perform the desired detection tasks is prohibitive-typically fifty thousand to hundreds of thousands of dollars. In a previous study at the Agricultural Research Organization’s Volcani Institute (Israel), a low-cost, high-performing HS system was developed which included a point spectrometer and optical components. Its main disadvantage was long shooting time for each image. Shooting time strongly depends on the predetermined integration time of the point spectrometer. While essential for performing monitoring tasks in a reasonable time, shortening integration time from a typical value in the range of 200 ms to the 10 ms range results in deterioration of the dynamic range of the captured scene. In this work, we suggest correcting this by learning the transformation from data measured with short integration time to that measured with long integration time. Reduction of the dynamic range and consequent low SNR were successfully overcome using three developed deep neural networks models based on a denoising auto-encoder, DnCNN and LambdaNetworks architectures as a backbone. The best model was based on DnCNN using a combined loss function of ℓ2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\ell _{2}$$\\end{document} and Kullback–Leibler divergence on images with 20 consecutive channels. The full spectrum of the model achieved a mean PSNR of 30.61 and mean SSIM of 0.9, showing total improvement relatively to the 10 ms measurements’ mean PSNR and mean SSIM values by 60.43% and 94.51%, respectively.

Low-light images enhancement via a dense transformer network

This paper proposes a dense network composed of an improved Transformer network, which successfully restores low-light images to high-quality normal-light images, alleviating issues such as low brightness, high noise, and missing critical information in low-light images. The entire network architecture is based on the improved Transformer network and builds a dense network with a combination of long and short connections. While retaining the self-attention mechanism of the Transformer network, it achieves multi-level fusion and utilization of shallow and deep features, providing the network with rich image features and enabling the restoration of low-light images to high-quality normal-light images. Additionally, a spatial-domain and frequency-domain combined loss function is designed, considering both pixel-level and frequency domain losses, effectively constraining the image restoration process and avoiding spectral biases. Lastly, a multi-scale hybrid gate feedforward network is designed to replace the traditional feedforward network in the Transformer, facilitating feature selection and forward propagation. These designs effectively enhance the richness of meaningful image features, alleviate spectral biases, and improve the visual quality of low-light images. Experimental results demonstrate the superiority of our method over state-of-the-art networks on various typical image enhancement datasets. Taking the most commonly used low-light dataset LOLv1 as an example, our method achieves improvements of 1.3% and 3.07% in PSNR and SSIM, respectively, compared to the best-performing network, showing favorable qualitative and quantitative evaluation results. The proposed method effectively addresses the issue of insufficiently realistic results in low-light image restoration, providing a reliable reference for practical applications.

CAM based fine-grained spatial feature supervision for hierarchical yoga pose classification using multi-stage transfer learning

In this paper, we propose a technique for hierarchical yoga pose classification (YPC) in a multi-stage multi-tasking framework. We propose a three-stage transfer learning based end-to-end training methodology. Novelty lies in (a) proposed supervised contrastive combined loss function for stage-1 training, (b) proposed Encoder–Decoder network architecture with attention mechanism for stage-3 training, (c) proposed spatial context aware multi-tasking combined loss function for stage-3 training. Firstly, for stage-1 training, we propose the usage of linear combination of three loss functions: cross-entropy, self-supervised contrastive loss and supervised contrastive loss in a multi-tasking manner. We introduce radial and cosine margin into the formulation of self-supervised and supervised contrastive loss to pull feature embeddings of same class closer together compared to feature embeddings of different classes. Weights learned over stage-1 training are subsequently fine-tuned over cross-entropy multi-tasking loss in stage-2. These stage-2 weights are transfer learned and are further fine-tuned in stage-3 training. For stage-3 training, we propose the usage of spatial context aware multi-tasking combined loss function. This loss function leverages on the fine-grained spatial features obtained from HiResCAM. These are processed in parallel with features obtained from XGrad-CAM (sensitivity and conservation axioms satisfying features) to further supervise the learning of hierarchical yoga pose classifier. We exemplify our methodology on the publicly available Yoga-82 large-scale dataset. We report peak Top-1 YPC accuracy of 95.89% over 6 pose classes (Yoga-6), 93.85% over 20 pose classes (Yoga-20) and 90.0% over 82 pose classes (Yoga-82). Our proposed method achieves 6.1% improvement over Top-1 classification accuracy in Yoga-6 hierarchy, 9.3% improvement in Yoga-20 hierarchy and 10.9% improvement in Yoga-82 hierarchy in comparison with state-of-the-art (SOTA) methodology. We achieve the current best Top-1 classification accuracies in all the three YPC hierarchies.

Self-equilibrium segmentation of near-infrared images of dental microcracks

The issue of missing teeth caused by dental microcracks has become a central topic in dental clinical medicine. Near-infrared technology combined with advanced image segmentation algorithms has become one of the leading technical routes for dental clinical medicine diagnosis. In this study, the NIR image segmentation method of dental microcracks was studied in depth to further utilize the advantages of near-infrared imaging technology in the detection technology of dental microcracks. The main contributions and research contents of this paper include (1) we designed a near-infrared imaging system for dental imaging, which provides solid data support for the study of advanced segmentation algorithms; (2) we designed a self-equilibrium segmentation algorithm for occult fracture of teeth based on U-Net++. The core module of this algorithm includes a multi-resolution residual encoder and a self-equilibrium composite loss function. The multi-scale residual encoder significantly improves the feature extraction capability of U-Net++. The self-equilibrium combined loss function ensures the consistency of the model’s attention to positive and negative samples, which enhances the robustness of U-Net++. (3) We have demonstrated through extensive experiments that our algorithm performs better than other dental occultation segmentation methods. This study is valuable for improving the diagnostic accuracy of dental occult fractures and promptly detecting the potential danger of the dental.

Improved Detection Method for Micro-Targets in Remote Sensing Images

With the exponential growth of remote sensing images in recent years, there has been a significant increase in demand for micro-target detection. Recently, effective detection methods for small targets have emerged; however, for micro-targets (even fewer pixels than small targets), most existing methods are not fully competent in feature extraction, target positioning, and rapid classification. This study proposes an enhanced detection method, especially for micro-targets, in which a combined loss function (consisting of NWD and CIOU) is used instead of a singular CIOU loss function. In addition, the lightweight Content-Aware Reassembly of Features (CARAFE) replaces the original bilinear interpolation upsampling algorithm, and a spatial pyramid structure is added into the network model’s small target layer. The proposed algorithm undergoes training and validation utilizing the benchmark dataset known as AI-TOD. Compared to speed-oriented YOLOv7-tiny, the mAP0.5 and mAP0.5:0.95 of our improved algorithm increased from 42.0% and 16.8% to 48.7% and 18.9%, representing improvements of 6.7% and 2.1%, respectively, while the detection speed was almost equal to that of YOLOv7-tiny. Furthermore, our method was also tested on a dataset of multi-scale targets, which contains small targets, medium targets, and large targets. The results demonstrated that mAP0.5:0.95 increased from “9.8%, 54.8%, and 68.2%” to “12.6%, 55.6%, and 70.1%” for detection across different scales, indicating improvements of 2.8%, 0.8%, and 1.9%, respectively. In summary, the presented method improves detection metrics for micro-targets in various scenarios while satisfying the requirements of detection speed in a real-time system.

Wide-field color imaging through multimode fiber with single wavelength illumination: plug-and-play approach.

Multimode fiber (MMF) is extensively studied for its ability to transmit light modes in parallel, potentially minimizing optical fiber size in imaging. However, current research predominantly focuses on grayscale imaging, with limited attention to color studies. Existing colorization methods often involve costly white light lasers or multiple light sources, increasing optical system expenses and space. To achieve wide-field color images with typical monochromatic illumination MMF imaging system, we proposed a data-driven "colorization" approach and a neural network called SpeckleColorNet, merging U-Net and conditional GAN (cGAN) architectures, trained by a combined loss function. This approach, demonstrated on a 2-meter MMF system with single-wavelength illumination and the Peripheral Blood Cell (PBC) dataset, outperforms grayscale imaging and alternative colorization methods in readability, definition, detail, and accuracy. Our method aims to integrate MMF into clinical medicine and industrial monitoring, offering cost-effective high-fidelity color imaging. It serves as a plug-and-play replacement for conventional grayscale algorithms in MMF systems, eliminating the need for additional hardware.

Joint MAPLE: Accelerated joint T1 and mapping with scan-specific self-supervised networks.

Quantitative MRI finds important applications in clinical and research studies. However, it is encoding intensive and may suffer from prohibitively long scan times. Accelerated MR parameter mapping techniques have been developed to help address these challenges. Here, an accelerated joint T1, , frequency and proton density mapping technique with scan-specific self-supervised network reconstruction is proposed to synergistically combine parallel imaging, model-based, and deep learning approaches to speed up parameter mapping. Proposed framework, Joint MAPLE, includes parallel imaging, signal modeling, and data consistency blocks which are optimized jointly in a combined loss function. A scan-specific self-supervised reconstruction is embedded into the framework, which takes advantage of multi-contrast data from a multi-echo, multi-flip angle, gradient echo acquisition. In comparison with parallel reconstruction techniques powered by low-rank methods, emerging scan specific networks, and model-based estimation approaches, the proposed framework reduces the reconstruction error in parameter maps by approximately two-fold on average at acceleration rates as high as R = 16 with uniform sampling. It can outperform evaluated parallel reconstruction techniques up to four-fold on average in the presence of challenging sub-sampling masks. It is observed that Joint MAPLE performs well at extreme acceleration rates of R = 25 and R = 36 with error values less than 20%. Joint MAPLE enables higher fidelity parameter estimation at high acceleration rates by synergistically combining parallel imaging and model-based parameter mapping and exploiting multi-echo, multi-flip angle datasets. Utilizing a scan-specific self-supervised reconstruction obviates the need for large data sets for training while improving the parameter estimation ability.