Search

Practical Application Apples are susceptible to postharvest bruises, leading to a shortened shelf life and significant waste. Therefore, accurate detection of apple bruises is crucial to mitigate food waste. This study proposed an improved lightweight network based on MobileViT for detecting early‐stage bruises in apples, utilizing hyperspectral imaging technology from 397.66 to 1003.81 nm. After acquiring hyperspectral images, the Otsu threshold algorithm was employed for mask extraction, and principal component analysis was used for feature image extraction. Subsequently, the improved MobileViT network (iM‐ViT) was implemented and compared with traditional algorithms, utilizing depthwise separable convolutions for parameter reduction and integrating local and global features to enhance bruise detection capability. The results demonstrated the superior performance of iM‐ViT in accurately detecting apple bruises, showing significant improvements. The F1 score and test accuracy for detecting apple bruises using iM‐ViT reached 0.99 and 99.07%, respectively. The fivefold cross‐validation strategy was used to assess the stability and robustness of iM‐ViT, and ablation experiments were performed to explore the effects of depthwise separable convolutions and local features on parameter reduction and classification accuracy improvement for early‐stage bruise detection in apples. The results demonstrated that iM‐ViT effectively reduced parameters and improved the ability to detect early bruises in apples.This study proposed an improved lightweight network to detect early bruises in apples, providing a reference for quick detection of bruises caused in the production process. Potential insights into the nondestructive detection of apple bruises using lightweight networks have been presented, which might be applied to mobile or online devices. [ABSTRACT FROM AUTHOR]

As image manipulation technology advances rapidly, the malicious use of image tampering has alarmingly escalated, posing a significant threat to social stability. In the realm of image tampering localization, accurately localizing limited samples, multiple types, and various sizes of regions remains a multitude of challenges. These issues impede the model's universality and generalization capability and detrimentally affect its performance. To tackle these issues, we propose FL-MobileViT-an improved MobileViT model devised for image tampering localization. Our proposed model utilizes a dual-stream architecture that independently processes the RGB and noise domain, and captures richer traces of tampering through dual-stream integration. Meanwhile, the model incorporating the Focused Linear Attention mechanism within the lightweight network (MobileViT). This substitution significantly diminishes computational complexity and resolves homogeneity problems associated with traditional Transformer attention mechanisms, enhancing feature extraction diversity and improving the model's localization performance. To comprehensively fuse the generated results from both feature extractors, we introduce the ASPP architecture for multi-scale feature fusion. This facilitates a more precise localization of tampered regions of various sizes. Furthermore, to bolster the model's generalization ability, we adopt a contrastive learning method and devise a joint optimization training strategy that leverages fused features and captures the disparities in feature distribution in tampered images. This strategy enables the learning of contrastive loss at various stages of the feature extractor and employs it as an additional constraint condition in conjunction with cross-entropy loss. As a result, overfitting issues are effectively alleviated, and the differentiation between tampered and untampered regions is enhanced. Experimental evaluations on five benchmark datasets (IMD-20, CASIA, NIST-16, Columbia and Coverage) validate the effectiveness of our proposed model. The meticulously calibrated FL-MobileViT model consistently outperforms numerous existing general models regarding localization accuracy across diverse datasets, demonstrating superior adaptability. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Railway Science & Engineering is the property of Journal of Railway Science & Engineering Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Color image binarization plays a pivotal role in image preprocessing work and significantly impacts subsequent tasks, particularly for text recognition. This paper concentrates on document image binarization (DIB), which aims to separate an image into a foreground (text) and background (non-text content). We thoroughly analyze conventional and deep-learning-based approaches and conclude that prevailing DIB methods leverage deep learning technology. Furthermore, we explore the receptive fields of pre- and post-network training to underscore the Transformer model's advantages. Subsequently, we introduce a lightweight model based on the U-Net structure and enhanced with the MobileViT module to capture global information features in document images better. Given its adeptness at learning both local and global features, our proposed model demonstrates competitive performance on two standard datasets (DIBCO2012 and DIBCO2017) and good robustness on the DIBCO2019 dataset. Notably, our proposed method presents a straightforward end-to-end model devoid of additional image preprocessing or post-processing, eschewing the use of ensemble models. Moreover, its parameter count is less than one-eighth of the model, which achieves the best results on most DIBCO datasets. Finally, two sets of ablation experiments are conducted to verify the effectiveness of the proposed binarization model. [ABSTRACT FROM AUTHOR]

Recent advancements in applications of deep neural network for bearing fault diagnosis under variable operating conditions have shown promising outcomes. However, these approaches are limited in practical applications due to the complexity of neural networks, which require substantial computational resources, thereby hindering the advancement of automated diagnostic tools. To overcome these limitations, this study introduces a new fault diagnosis framework that incorporates a tri-channel preprocessing module for multidimensional feature extraction, coupled with an innovative diagnostic architecture known as the Lightweight Ghost Enhanced Feature Attention Network (GEFA-Net). This system is adept at identifying rolling bearing faults across diverse operational conditions. The FFE module utilizes advanced techniques such as Fast Fourier Transform (FFT), Frequency Weighted Energy Operator (FWEO), and Signal Envelope Analysis to refine signal processing in complex environments. Concurrently, GEFA-Net employs the Ghost Module and the Efficient Pyramid Squared Attention (EPSA) mechanism, which enhances feature representation and generates additional feature maps through linear operations, thereby reducing computational demands. This methodology not only significantly lowers the parameter count of the model, promoting a more streamlined architectural framework, but also improves diagnostic speed. Additionally, the model exhibits enhanced diagnostic accuracy in challenging conditions through the effective synthesis of local and global data contexts. Experimental validation using datasets from the University of Ottawa and our dataset confirms that the framework not only achieves superior diagnostic accuracy but also reduces computational complexity and accelerates detection processes. These findings highlight the robustness of the framework for bearing fault diagnosis under varying operational conditions, showcasing its broad applicational potential in industrial settings. The parameter count was decreased by 63.74% compared to MobileVit, and the recorded diagnostic accuracies were 98.53% and 99.98% for the respective datasets. [ABSTRACT FROM AUTHOR]

From the perspective of lightweight models, a facial expression recognition network based on network fusion to improve Mobile ViT is proposed. This network integrates multiscale convolution PSConv and attention mechanisms through residual structures to form the RAPsconv feature reconstruction module. This module can more efficiently extract multiscale features from a fine-grained perspective, enhancing the expression of key features, thereby improving the network's expressive ability and constructing an end-to-end facial expression recognition network. Additionally, to further narrow the gap between similar expressions, a loss function combining Soft-max Loss and Center Loss is proposed, effectively reducing the misjudgment rate of expression recognition. Experimental results demonstrate that the improved network achieves higher accuracy on three natural scene expression datasets FER2013, FER+, and RAFDB compared to the base network Mobile ViT, with accuracy improvements of 1. 73%, 2. 18%, and 1. 64%, respectively. The improved network has fewer parameters, stronger robustness, and is suitable for light-weighting and integration, making it suitable for real-world applications in facial expression recognition. [ABSTRACT FROM AUTHOR]

To avoid the overfitting problem of the network model and improve the facial expression recognition effect of partially occluded facial images, an improved facial expression recognition algorithm based on MobileViT has been proposed. Firstly, in order to obtain features that are useful and richer for experiments, deep convolution operations are added to the inverted residual blocks of this network, thus improving the facial expression recognition rate. Then, in the process of dimension reduction, the activation function can significantly improve the convergence speed of the model, and then quickly reduce the loss error in the training process, as well as to preserve the effective facial expression features as much as possible and reduce the overfitting problem. Experimental results on RaFD, FER2013, and FER2013Plus show that this method has significant advantages over mainstream networks and the network achieves the highest recognition rate. [ABSTRACT FROM AUTHOR]

The long-distance link between facial landmarks cannot be modeled by the current CNN-based facial landmark detection networks, and these networks typically have many parameters that consume substantial computational resources. This paper proposes a multi-scale lightweight facial landmark detection network with CNN and Transformer multi-branch parallelism. Based on MobileViT, the network incorporates MobileOne Block and simplified Ghost BottleNeck lightweight network structure. Compared to MobileViT on the WFLW dataset, the number of network parameters is reduced by 49.18%, the failure rate is reduced by 3.20%, the detection speed is improved by 41.73%, the FLOPS is reduced by 64.83%, and the NME is improved by 0.45% and 1.31% on the test and pose subsets, respectively. The data proves that the global information extraction of facial landmarks is more accurate after adding the Transformer structure. This paper also compares with other networks, and the result shows that improved MobileViT achieves more accurate detection with fewer model parameters. [ABSTRACT FROM AUTHOR]

With a rapid development of artificial intelligence technology, fine-grained image classification has gained widespread application. For mobile terminals, this paper introduces an image classification method built on MobileViT, and it can apply into fine-grained image classification. The original MobileViT model has been optimized in three ways. Initially, the h-swish activation function is used to enhance the network performance. Second, the cross-entropy loss function is used to further realize the parameter optimization and model accuracy improvement. Finally, a dropout layer is joined before the fully connected layer can effectively decrease the model recognition time and prevent over-fitting. Experimental data on public tomato disease datasets demonstrate that the improved fine-grained image classification method put forward in this paper exhibits higher classification accuracy, better stability and network generalization ability than other models. [ABSTRACT FROM AUTHOR]

Software defect prediction is an important part of software development, which aims to use existing historical data to predict future software defects. Focusing on the model performance and communication efficiency of cross-project software defect prediction, this paper proposes an efficient communication-based federated meta-learning (ECFML) algorithm. The lightweight MobileViT network is used as the meta-learner of the Model Agnostic Meta-Learning (MAML) algorithm. By learning common knowledge on the local data of multiple clients, and then fine-tuning the model, the number of unnecessary iterations is reduced, and communication efficiency is improved while reducing the number of parameters. The gradient information model is encrypted using the differential privacy of the Laplace mechanism, and the optimal privacy budget is determined through experiments. Experiments on three public datasets (AEEEM, NASA, and Relink) verified the effectiveness of ECFML in terms of parameter quantity, convergence, and model performance of cross-project software defect prediction. [ABSTRACT FROM AUTHOR]

The tea industry, as one of the most globally important agricultural products, is characterized by pests and diseases that pose a serious threat to yield and quality. These diseases and pests often present different scales and morphologies, and some pest and disease target sizes can be tiny and difficult to detect. To solve these problems, we propose TeaViTNet, a multi-scale attention-based tea pest and disease detection model that combines CNNs and Transformers. First, MobileViT is used as the feature extraction backbone network. MobileViT captures and analyzes the tiny pest and disease features in the image via a self-attention mechanism and global feature extraction. Second, the EMA-PANet network is introduced to optimize the model's learning and attention to the Apolygus lucorum and leaf blight regions via an efficient multi-scale attention module with cross-space learning, which improves the model's ability to understand multi-scale information. In addition, RFBNet is embedded in the module to further expand the perceptual range and effectively capture the information of tiny features in tea leaf images. Finally, the ODCSPLayer convolutional block is introduced, aiming to focus on acquiring richer gradient flow information. The experimental results show that the TeaViTNet model proposed in this paper has an average accuracy of 89.1%, which is a significant improvement over the baseline network MobileViT and is capable of accurately detecting Apolygus lucorum and leaf blight of different scales and complexities. [ABSTRACT FROM AUTHOR]

Mild cognitive impairment (MCI) is an early stage of Alzheimer’s disease (AD), which is currently incurable. Early diagnosis of AD is essential for effective intervention since the World Alzheimer’s Report 2015 predicted the number of cases will triple by 2050. The 18F-FDG PET imaging technique, although effective in detecting metabolic activities in the brain, faces challenges such as low signal-to-noise ratios and limited data availability, which complicates the extraction of necessary lesion information that is effective for early stage MCI diagnose. To overcome these challenges, we introduce a novel deep learning-based model, ResGLPyramid, that combines convolution operations, MobileViTv3, and a global-local attention module (GLAM) block, to capture local and global representations. By utilizing a softened cross-entropy (SCE) objective function, the model reduces overfitting, improves generalization, and enhances the detection of subtle metabolic changes. The proposed model enhances the sensitivity and specificity of Alzheimer’s detection by leveraging local- and long-range interactions among critical diagnostic features that lead to more precise and efficient analyses. The experimental results show that the ResGLPyramid model achieved an accuracy of 92.75%, sensitivity of 90.80%, and specificity of 94.14% in classifying MCI and AD individuals. These results represent a 3.44% increase in accuracy and a 4.28% increase in specificity while the sensitivity is slightly lower compared to state-of-the-art methods.

As dust source bare soil land (BSL) contributes to air pollution and affects the photosynthesis of green plants and carbon absorption, it is the objective of this study to develop an approach for monitoring the changes of BSL using remote sensing technology. Unlike other land use/cover types, the classification of BSL as well as its change detection is often ignored. For traditional convolutional neural networks, deep layers cause a long range between input and output, inevitably leading to the loss of information and computational costs. To alleviate this problem, transformer is available to model the global dependencies. Bitemporal association, which is described as subtraction or attention mechanism, is not fully considered by current methods. Therefore, we proposed a spatial-temporal association enhanced mobile-friendly vision transformer (STAE-MobileVIT) for change detection of high-resolution images with light weight and high efficiency. On the one hand, a temporal association enhanced MobileVIT block is employed to strengthen the association of bitemporal images during feature extraction. On the other hand, a multiscale feature difference aggregator enhanced by spatial association is designed to fuse semantic and detailed information. Since the lack of binary change detection dataset for BSL, we established a small dataset named BSL-CD, consisting of 1083 pairs of 0.8 m bitemporal images with the size of 256 × 256 pixels, along with the corresponding labels. The experiments on BSL-CD show that our light-weight model surpass seven common methods by 3.48, 5.05, and 1.44 percent on F1, IoU, and OA, which proves the efficiency and accuracy of STAE-MobileVIT.

The presence of corn unsound kernels, one of the world's major food crops, could have a significant impact on the food industry and the food supply. The traditional method of detecting unsound corn kernels by hand during grain acquisition has many drawbacks, and computer vision‐based detection methods have become mainstream. In this paper, a corn unsound kernel detection algorithm based on generative adversarial network (GAN) sample enhancement and an improved lightweight network is introduced. The article first builds a corn unsound kernel image acquisition platform and makes a dataset by preprocessing and segmenting the collected corn seed cluster images with the improved concave point segmentation algorithm. Then, to increase the diversity and number of datasets, the StyleGANv2 network was improved to generate corn‐unsound kernel images with diverse features. Finally, to meet the demand for lightweight detection, the MobileVit network was optimized to improve the network's recognition accuracy, which reached 96.2%. The article verifies the effectiveness and superiority of the proposed algorithm through experiments. [ABSTRACT FROM AUTHOR]

As the sheep industry rapidly moves towards modernization, digitization, and intelligence, there is a need to build breeding farms integrated with big data. By collecting individual information on sheep, precision breeding can be conducted to improve breeding efficiency, reduce costs, and promote healthy breeding practices. In this context, the accurate identification of individual sheep is essential for establishing digitized sheep farms and precision animal husbandry. Currently, scholars utilize deep learning technology to construct recognition models, learning the biological features of sheep faces to achieve accurate identification. However, existing research methods are limited to pattern recognition at the image level, leading to a lack of diversity in recognition methods. Therefore, this study focuses on the small-tailed Han sheep and develops a sheep face recognition method based on three-dimensional reconstruction technology and feature point matching, aiming to enrich the theoretical research of sheep face recognition technology. The specific recognition approach is as follows: full-angle sheep face images of experimental sheep are collected, and corresponding three-dimensional sheep face models are generated using three-dimensional reconstruction technology, further obtaining three-dimensional sheep face images from three different perspectives. Additionally, this study developed a sheep face orientation recognition algorithm called the sheep face orientation recognition algorithm (SFORA). The SFORA incorporates the ECA mechanism to further enhance recognition performance. Ultimately, the SFORA has a model size of only 5.3 MB, with accuracy and F1 score reaching 99.6% and 99.5%, respectively. During the recognition task, the SFORA is first used for sheep face orientation recognition, followed by matching the recognition image with the corresponding three-dimensional sheep face image based on the established SuperGlue feature-matching algorithm, ultimately outputting the recognition result. Experimental results indicate that when the confidence threshold is set to 0.4, SuperGlue achieves the best matching performance, with matching accuracies for the front, left, and right faces reaching 96.0%, 94.2%, and 96.3%, respectively. This study enriches the theoretical research on sheep face recognition technology and provides technical support.

Recent advancements in applications of deep neural network for bearing fault diagnosis under variable operating conditions have shown promising outcomes. However, these approaches are limited in practical applications due to the complexity of neural networks, which require substantial computational resources, thereby hindering the advancement of automated diagnostic tools. To overcome these limitations, this study introduces a new fault diagnosis framework that incorporates a tri-channel preprocessing module for multidimensional feature extraction, coupled with an innovative diagnostic architecture known as the Lightweight Ghost Enhanced Feature Attention Network (GEFA-Net). This system is adept at identifying rolling bearing faults across diverse operational conditions. The FFE module utilizes advanced techniques such as Fast Fourier Transform (FFT), Frequency Weighted Energy Operator (FWEO), and Signal Envelope Analysis to refine signal processing in complex environments. Concurrently, GEFA-Net employs the Ghost Module and the Efficient Pyramid Squared Attention (EPSA) mechanism, which enhances feature representation and generates additional feature maps through linear operations, thereby reducing computational demands. This methodology not only significantly lowers the parameter count of the model, promoting a more streamlined architectural framework, but also improves diagnostic speed. Additionally, the model exhibits enhanced diagnostic accuracy in challenging conditions through the effective synthesis of local and global data contexts. Experimental validation using datasets from the University of Ottawa and our dataset confirms that the framework not only achieves superior diagnostic accuracy but also reduces computational complexity and accelerates detection processes. These findings highlight the robustness of the framework for bearing fault diagnosis under varying operational conditions, showcasing its broad applicational potential in industrial settings. The parameter count was decreased by 63.74% compared to MobileVit, and the recorded diagnostic accuracies were 98.53% and 99.98% for the respective datasets.

Abstract Background The medical profession is facing an excessive workload, which has led to the development of various Computer-Aided Diagnosis (CAD) systems as well as Mobile-Aid Diagnosis (MAD) systems. These technologies enhance the speed and accuracy of diagnoses, particularly in areas with limited resources or remote regions during the pandemic. The primary purpose of this research is to predict and diagnose COVID-19 infection from chest X-ray images by developing a mobile-friendly deep learning framework, which has the potential for deployment in portable devices such as mobile or tablet, especially in situations where the workload of radiology specialists may be high. Moreover, this could improve the accuracy and transparency of population screening to assist radiologists during the pandemic. Methods In this study, the Mobile Networks ensemble model called COV-MobNets is proposed to classify positive COVID-19 X-ray images from negative ones and can have an assistant role in diagnosing COVID-19. The proposed model is an ensemble model, combining two lightweight and mobile-friendly models: MobileViT based on transformer structure and MobileNetV3 based on Convolutional Neural Network. Hence, COV-MobNets can extract the features of chest X-ray images in two different methods to achieve better and more accurate results. In addition, data augmentation techniques were applied to the dataset to avoid overfitting during the training process. The COVIDx-CXR-3 benchmark dataset was used for training and evaluation. Results The classification accuracy of the improved MobileViT and MobileNetV3 models on the test set has reached 92.5% and 97%, respectively, while the accuracy of the proposed model (COV-MobNets) has reached 97.75%. The sensitivity and specificity of the proposed model have also reached 98.5% and 97%, respectively. Experimental comparison proves the result is more accurate and balanced than other methods. Conclusion The proposed method can distinguish between positive and negative COVID-19 cases more accurately and quickly. The proposed method proves that utilizing two automatic feature extractors with different structures as an overall framework of COVID-19 diagnosis can lead to improved performance, enhanced accuracy, and better generalization to new or unseen data. As a result, the proposed framework in this study can be used as an effective method for computer-aided diagnosis and mobile-aided diagnosis of COVID-19. The code is available publicly for open access at https://github.com/MAmirEshraghi/COV-MobNets .

In tasks that require ship detection and recognition, the irregular shapes of ships and complex backgrounds pose significant challenges. This paper presents an advanced extension of the YOLOv8 model to address these challenges. A lightweight visual transformer, MobileViTSF, is proposed and combined with the YOLOv8 model. To address the loss of semantic information that arises from inconsistent scales in the detection of small ships, a layer intended for the detection of small targets is introduced to lead to improved fusion of deep and shallow features. Furthermore, the traditional convolution (Conv) blocks are replaced with GSConv blocks, and a novel GSC2f block is designed for fewer model parameters and improved detection performance. Experiments on a benchmark dataset suggest that this new model can achieve significantly improved accuracy for ship detection with fewer model parameters and a reduced model size. A comparison with several other state-of-the-art methods shows that higher accuracy can be obtained for ship detection with this model. Moreover, this new model is suitable for edge computing devices, demonstrating practical application value. [ABSTRACT FROM AUTHOR]

Agricultural production utilizes pesticides as a crucial factor in protecting crops to supplement the food supply amidst today's increasing demand. Owing to significant financial and moral benefits, gardeners continue to tolerate pesticide usage, which, to a certain extent, can adversely affect farming operations. Our goal is to employ deep machine learning algorithms to detect pesticide misuse activities on farms using easily accessible surveillance cameras or video clips. This information could provide vital data for promoting the safe use of vegetables and pesticides for consumers, as well as enabling authorities to swiftly evaluate the quality of agricultural products. We developed PesViT as our primary model, which is based on end-to-end convolutional model optimization in MobileViT using Ghost blocks. We then applied contrastive self-supervised learning method with momentum contrast technique (SSL-MoCo). This process will utilize our unlabeled data sets with appropriate adjustments to the hyperparameters for the most accurate and fastest outputs. Current data collected on farms at various times of the day have shown that the model's efficiency reaches 95.36%, compared to the original MobileViT's 91.25% or MobileNetV2's 88.75%. Notably, the experimental build performs well on limited-configuration computers, suggesting a promising future for the widespread application of the model. [ABSTRACT FROM AUTHOR]

Copyright of Experimental Technology & Management is the property of Experimental Technology & Management Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

To avoid the overfitting problem of the network model and improve the facial expression recognition effect of partially occluded facial images, an improved facial expression recognition algorithm based on MobileViT has been proposed. Firstly, in order to obtain features that are useful and richer for experiments, deep convolution operations are added to the inverted residual blocks of this network, thus improving the facial expression recognition rate. Then, in the process of dimension reduction, the activation function can significantly improve the convergence speed of the model, and then quickly reduce the loss error in the training process, as well as to preserve the effective facial expression features as much as possible and reduce the overfitting problem. Experimental results on RaFD, FER2013, and FER2013Plus show that this method has significant advantages over mainstream networks and the network achieves the highest recognition rate.

The tea industry, as one of the most globally important agricultural products, is characterized by pests and diseases that pose a serious threat to yield and quality. These diseases and pests often present different scales and morphologies, and some pest and disease target sizes can be tiny and difficult to detect. To solve these problems, we propose TeaViTNet, a multi-scale attention-based tea pest and disease detection model that combines CNNs and Transformers. First, MobileViT is used as the feature extraction backbone network. MobileViT captures and analyzes the tiny pest and disease features in the image via a self-attention mechanism and global feature extraction. Second, the EMA-PANet network is introduced to optimize the model’s learning and attention to the Apolygus lucorum and leaf blight regions via an efficient multi-scale attention module with cross-space learning, which improves the model’s ability to understand multi-scale information. In addition, RFBNet is embedded in the module to further expand the perceptual range and effectively capture the information of tiny features in tea leaf images. Finally, the ODCSPLayer convolutional block is introduced, aiming to focus on acquiring richer gradient flow information. The experimental results show that the TeaViTNet model proposed in this paper has an average accuracy of 89.1%, which is a significant improvement over the baseline network MobileViT and is capable of accurately detecting Apolygus lucorum and leaf blight of different scales and complexities.

Image segmentation has played an essential role in computer vision. The target detection model represented by YOLOv5 is widely used in image segmentation. However, YOLOv5 has performance bottlenecks such as object scale variation, object occlusion, computational volume, and speed when processing complex images. To solve these problems, an enhanced algorithm based on YOLOv5 is proposed. MobileViT is used as the backbone network of the YOLOv5 algorithm, and feature fusion and dilated convolution are added to the model. This method is validated on the COCO and PASCAL-VOC datasets. Experimental results show that it significantly reduces the processing time and achieves high segmentation quality with an accuracy of 95.32% on COCO and 96.02% on PASCAL-VOC. The improved model is 116 M, 52 M, and 76 M, smaller than U-Net, SegNet, and Mask R-CNN, respectively. This paper provides a new idea and method with which to solve the problems in the field of image segmentation, and the method has strong practicality and generalization value. [ABSTRACT FROM AUTHOR]

BACKGROUND: Accurate classification of benign and malignant pulmonary nodules using chest computed tomography (CT) images is important for early diagnosis and treatment of lung cancer. In terms of natural image classification, the ViT-based model has greater advantages in extracting global features than the traditional CNN model. However, due to the small image dataset and low image resolution, it is difficult to directly apply the ViT-based model to pulmonary nodule classification. OBJECTIVE: To propose and test a new ViT-based MSM-ViT model aiming to achieve good performance in classifying pulmonary nodules. METHODS: In this study, CNN structure was used in the task of classifying pulmonary nodules to compensate for the poor generalization of ViT structure and the difficulty in extracting multi-scale features. First, sub-pixel fusion was designed to improve the ability of the model to extract tiny features. Second, multi-scale local features were extracted by combining dilated convolution with ordinary convolution. Finally, MobileViT module was used to extract global features and predict them at the spatial level. RESULTS: CT images involving 442 benign nodules and 406 malignant nodules were extracted from LIDC-IDRI data set to verify model performance, which yielded the best accuracy of 94.04% and AUC value of 0.9636 after 10 cross-validations. CONCLUSION: The proposed new model can effectively extract multi-scale local and global features. The new model performance is also comparable to the most advanced models that use 3D volume data training, but its occupation of video memory (training resources) is less than 1/10 of the conventional 3D models. [ABSTRACT FROM AUTHOR]

Background: The medical profession is facing an excessive workload, which has led to the development of various Computer-Aided Diagnosis (CAD) systems as well as Mobile-Aid Diagnosis (MAD) systems. These technologies enhance the speed and accuracy of diagnoses, particularly in areas with limited resources or remote regions during the pandemic. The primary purpose of this research is to predict and diagnose COVID-19 infection from chest X-ray images by developing a mobile-friendly deep learning framework, which has the potential for deployment in portable devices such as mobile or tablet, especially in situations where the workload of radiology specialists may be high. Moreover, this could improve the accuracy and transparency of population screening to assist radiologists during the pandemic. Methods: In this study, the Mobile Networks ensemble model called COV-MobNets is proposed to classify positive COVID-19 X-ray images from negative ones and can have an assistant role in diagnosing COVID-19. The proposed model is an ensemble model, combining two lightweight and mobile-friendly models: MobileViT based on transformer structure and MobileNetV3 based on Convolutional Neural Network. Hence, COV-MobNets can extract the features of chest X-ray images in two different methods to achieve better and more accurate results. In addition, data augmentation techniques were applied to the dataset to avoid overfitting during the training process. The COVIDx-CXR-3 benchmark dataset was used for training and evaluation. Results: The classification accuracy of the improved MobileViT and MobileNetV3 models on the test set has reached 92.5% and 97%, respectively, while the accuracy of the proposed model (COV-MobNets) has reached 97.75%. The sensitivity and specificity of the proposed model have also reached 98.5% and 97%, respectively. Experimental comparison proves the result is more accurate and balanced than other methods. Conclusion: The proposed method can distinguish between positive and negative COVID-19 cases more accurately and quickly. The proposed method proves that utilizing two automatic feature extractors with different structures as an overall framework of COVID-19 diagnosis can lead to improved performance, enhanced accuracy, and better generalization to new or unseen data. As a result, the proposed framework in this study can be used as an effective method for computer-aided diagnosis and mobile-aided diagnosis of COVID-19. The code is available publicly for open access at https://github.com/MAmirEshraghi/COV-MobNets. [ABSTRACT FROM AUTHOR]

The underwater imaging environment is complex, and the application of conventional target detection algorithms to the underwater environment has yet to provide satisfactory results. Therefore, underwater optical image target detection remains one of the most challenging tasks involved with neighborhood-based techniques in the field of computer vision. Small underwater targets, dispersion, and sources of distortion (such as sediment and particles) often render neighborhood-based techniques insufficient, as existing target detection algorithms primarily focus on improving detection accuracy and enhancing algorithm complexity and computing power. However, excessive extraction of deep-level features leads to the loss of small targets and decrease in detection accuracy. Moreover, most underwater optical image target detection is performed by underwater unmanned platforms, which have a high demand of algorithm lightweight requirements due to the limited computing power of the underwater unmanned platform with the mobile vision processing platform. In order to meet the lightweight requirements of the underwater unmanned platform without affecting the detection accuracy of the target, we propose an underwater target detection model based on mobile vision transformer (MobileViT) and YOLOX, and we design a new coordinate attention (CA) mechanism named a double CA (DCA) mechanism. This model utilizes MobileViT as the algorithm backbone network, improving the global feature extraction ability of the algorithm and reducing the amount of algorithm parameters. The double CA (DCA) mechanism can improve the extraction of shallow features as well as the detection accuracy, even for difficult targets, using a minimum of parameters. Research validated in the Underwater Robot Professional Contest 2020 (URPC2020) dataset revealed that this method has an average accuracy rate of 72.00%. In addition, YOLOX's ability to compress the model parameters by 49.6% efficiently achieves a balance between underwater optical image detection accuracy and parameter quantity. Compared with the existing algorithm, the proposed algorithm can carry on the underwater unmanned platform better. [ABSTRACT FROM AUTHOR]

Weeds in the field are serious threats to agricultural production. The rapid and accurate identification of weed species can be critical steps to automated weed control. However, previous weed recognition relies mainly on hand-designed features, such as shape and texture. Fortunately, computer vision can be expected to serve as intelligent weed control equipment. The performance can also be obtained better on the data samples with significant morphological differences. But it cannot accurately identify the weeds and crops in field environments. It is highly required to promote the expression capabilities of the weed species during recognition in real time, due to the particularity of field weeding tasks. In this study, a novel weed recognition was proposed to balance the high accuracy and real-time performance using the improved lightweight MobileViT network. The efficient Channel Attention (ECA) module was also combined to enhance the important parts of the features for better extraction. Firstly, the MobileViT network structure was selected as the lightweight model using Transformers. Specifically, the MobileViT backbone network was used to replace the self-attention layer in the ViT model with the lightweight separable convolution layer. Some optimization techniques were also introduced, such as the depth separable convolution and channel attention mechanism. The feature extraction was then achieved at a high inference speed, but a lower model size and computational cost. As such, the core network was established for lightweight weed identification. Secondly, the ECA mechanism was used as the learnable channel weights to calculate the importance of each channel, with emphasis on the important channels in the input features. The important features were enhanced to further improve the extraction capability from the down-sampled map. Finally, the MobileViT module was trained to simultaneously learn both local and global semantic information in the feature extraction network using both MobileViT and MobileNet networks. Only a small number of modules were combined to accurately capture the subtle differences between the different classes of weeds and crops, in order to fully meet the requirements of real-time performance and recognition accuracy. The comparative experiments were conducted to verify the effectiveness of the model. The images of corn seedlings and four types of companion weeds were also collected in the field environment. The results show that the recognition accuracy, precision, recall, and F1 score of the improved model were 99.61%, 99.60%, 99.58%, and 99.59%, respectively, indicating better performance than the lightweight convolutional neural networks (CNNs), such as ShuffleNet and MobileNet, and the general CNNs, such as VGG-16, ResNet-50, and DenseNet-161. Furthermore, the inference time was only 83ms for the single image, indicating that the recognition efficiency fully met the real-time requirement for the weed identification during weeding operations. The visualization also showed that the key features were effectively extracted from the weed images without the influence of background areas. The new model can be expected to accurately and rapidly distinguish the multiple morphologically similar weeds and crops in a field environment. The finding can also provide a strong reference for the weed identification systems in intelligent weed control equipment. [ABSTRACT FROM AUTHOR]

Wild mushrooms are not only tasty but also rich in nutritional value, but it is difficult for non-specialists to distinguish poisonous wild mushrooms accurately. Given the frequent occurrence of wild mushroom poisoning, we propose a new multidimensional feature fusion attention network (M-ViT) combining convolutional networks (ConvNets) and attention networks to compensate for the deficiency of pure ConvNets and pure attention networks. First, we introduced an attention mechanism Squeeze and Excitation (SE) module in the MobilenetV2 (MV2) structure of the network to enhance the representation of picture channels. Then, we designed a Multidimension Attention module (MDA) to guide the network to thoroughly learn and utilize local and global features through short connections. Moreover, using the Atrous Spatial Pyramid Pooling (ASPP) module to obtain longer distance relations, we fused the model features from different layers, and used the obtained joint features for wild mushroom classification. We validated the model on two datasets, mushroom and MO106, and the results showed that M-ViT performed the best on the two test datasets, with accurate dimensions of 96.21% and 91.83%, respectively. We compared the performance of our method with that of more advanced ConvNets and attention networks (Transformer), and our method achieved good results. [ABSTRACT FROM AUTHOR]

Recently, deep learning methods that use a combination of convolutional neural networks and Transformers have shown excellent results in both completely supervised and semi-supervised medical image segmentation tasks. This study provides a simple and effective framework for semi-supervised medical image segmentation by introducing cross teaching between MobileNet and MobileViT. Specifically, we built two different two-path parallel semantic segmentation networks with MobileNet and MobileViT as the main modules. Cross teaching between MobileNet and MobileViT was performed, in which the prediction of each network was used as a pseudo-label to supervise the training of other networks in a direct end-to-end manner. Finally, experiments on two common benchmark tests showed that the proposed framework was superior to six existing semi-supervised learning methods. This shows that our framework can effectively use unlabeled data to improve the performance and is superior to the latest semi-supervised segmentation method. Codes are available at: github.com/yywbkn/MV2-cross-teaching-MobileViT. • A simple and effective semi-supervised scheme, based on MobileNet and MobileViT. • A new cross teaching framework for medical image segmentation. • This framework performance reaches the level of the state-of-art semi-supervised scheme. [ABSTRACT FROM AUTHOR]

Image segmentation has played an essential role in computer vision. The target detection model represented by YOLOv5 is widely used in image segmentation. However, YOLOv5 has performance bottlenecks such as object scale variation, object occlusion, computational volume, and speed when processing complex images. To solve these problems, an enhanced algorithm based on YOLOv5 is proposed. MobileViT is used as the backbone network of the YOLOv5 algorithm, and feature fusion and dilated convolution are added to the model. This method is validated on the COCO and PASCAL-VOC datasets. Experimental results show that it significantly reduces the processing time and achieves high segmentation quality with an accuracy of 95.32% on COCO and 96.02% on PASCAL-VOC. The improved model is 116 M, 52 M, and 76 M, smaller than U-Net, SegNet, and Mask R-CNN, respectively. This paper provides a new idea and method with which to solve the problems in the field of image segmentation, and the method has strong practicality and generalization value.

The underwater imaging environment is complex, and the application of conventional target detection algorithms to the underwater environment has yet to provide satisfactory results. Therefore, underwater optical image target detection remains one of the most challenging tasks involved with neighborhood-based techniques in the field of computer vision. Small underwater targets, dispersion, and sources of distortion (such as sediment and particles) often render neighborhood-based techniques insufficient, as existing target detection algorithms primarily focus on improving detection accuracy and enhancing algorithm complexity and computing power. However, excessive extraction of deep-level features leads to the loss of small targets and decrease in detection accuracy. Moreover, most underwater optical image target detection is performed by underwater unmanned platforms, which have a high demand of algorithm lightweight requirements due to the limited computing power of the underwater unmanned platform with the mobile vision processing platform. In order to meet the lightweight requirements of the underwater unmanned platform without affecting the detection accuracy of the target, we propose an underwater target detection model based on mobile vision transformer (MobileViT) and YOLOX, and we design a new coordinate attention (CA) mechanism named a double CA (DCA) mechanism. This model utilizes MobileViT as the algorithm backbone network, improving the global feature extraction ability of the algorithm and reducing the amount of algorithm parameters. The double CA (DCA) mechanism can improve the extraction of shallow features as well as the detection accuracy, even for difficult targets, using a minimum of parameters. Research validated in the Underwater Robot Professional Contest 2020 (URPC2020) dataset revealed that this method has an average accuracy rate of 72.00%. In addition, YOLOX’s ability to compress the model parameters by 49.6% efficiently achieves a balance between underwater optical image detection accuracy and parameter quantity. Compared with the existing algorithm, the proposed algorithm can carry on the underwater unmanned platform better.

In view of the large amount of vehicle detection model parameters and missing detection of small targets and occluded targets. This paper presents a vehicle detection algorithm based on lightweight MobileVit. Firstly, in the data preprocessing stage, the method used GridMask image enhancement to improve the performance of occluded target detection; Secondly, the method used MobileVit network as the backbone feature extraction network of the model to fully extract the feature information and make the model lightweight; Finally, in the prediction layer network, the method used multiscale vehicle detection and recognition based on PANet network to improve the detection performance of the model for small targets. The experimental results show that the average detection accuracy of this algorithm is 98.24%, the detection speed is 58 ms per picture, and the model size is 136 MB, compared with the comparison algorithm, the comprehensive performance is better. [ABSTRACT FROM AUTHOR]

The soluble solids content (SSC) is an important indicator for evaluating the internal quality of apples. This study proposed a novel method for detecting apple SSC. This method involved collecting the transmission spectra of apples, transforming one-dimensional spectra into two-dimensional images using the gramian angular difference field (GADF) technique, and analyzing the images using an improved mobile vision transformer (MobileViT) model. Furthermore, the Grad-CAM method was used to visualize the model's prediction effect on SSC and find the wavelength range that had a greater impact on the model results. The results indicated that the GADF-improved MobileViT model exhibited excellent performance on the test dataset, with a determination coefficient (R 2) of 0.938, a root mean square error (RMSE) of 0.532, and a mean absolute error (MAE) of 0.423. As the number of wavelengths used for modeling decreased, the improvements made in this study to MobileViT continued to enhance the model's performance effectively. And when using fewer training samples, the GADF-improved MobileViT model also had better performance than the original model. In conclusion, using visible and near-infrared spectroscopy and GADF image encoding techniques for the prediction of apple SSC is feasible. • The SSC of golden venus apples was predicted by visible/NIR spectroscopy. • Converting 1D spectral data into 2D spectral images to predict apple SSC. • Proposed GADF-improved MobileViT based on transformer to predict apple SSC. • The Grad-CAM visualized influential wavelength ranges impacting model predictions. [ABSTRACT FROM AUTHOR]

Wild mushrooms are not only tasty but also rich in nutritional value, but it is difficult for non-specialists to distinguish poisonous wild mushrooms accurately. Given the frequent occurrence of wild mushroom poisoning, we propose a new multidimensional feature fusion attention network (M-ViT) combining convolutional networks (ConvNets) and attention networks to compensate for the deficiency of pure ConvNets and pure attention networks. First, we introduced an attention mechanism Squeeze and Excitation (SE) module in the MobilenetV2 (MV2) structure of the network to enhance the representation of picture channels. Then, we designed a Multidimension Attention module (MDA) to guide the network to thoroughly learn and utilize local and global features through short connections. Moreover, using the Atrous Spatial Pyramid Pooling (ASPP) module to obtain longer distance relations, we fused the model features from different layers, and used the obtained joint features for wild mushroom classification. We validated the model on two datasets, mushroom and MO106, and the results showed that M-ViT performed the best on the two test datasets, with accurate dimensions of 96.21% and 91.83%, respectively. We compared the performance of our method with that of more advanced ConvNets and attention networks (Transformer), and our method achieved good results.

As the sheep industry rapidly moves towards modernization, digitization, and intelligence, there is a need to build breeding farms integrated with big data. By collecting individual information on sheep, precision breeding can be conducted to improve breeding efficiency, reduce costs, and promote healthy breeding practices. In this context, the accurate identification of individual sheep is essential for establishing digitized sheep farms and precision animal husbandry. Currently, scholars utilize deep learning technology to construct recognition models, learning the biological features of sheep faces to achieve accurate identification. However, existing research methods are limited to pattern recognition at the image level, leading to a lack of diversity in recognition methods. Therefore, this study focuses on the small-tailed Han sheep and develops a sheep face recognition method based on three-dimensional reconstruction technology and feature point matching, aiming to enrich the theoretical research of sheep face recognition technology. The specific recognition approach is as follows: full-angle sheep face images of experimental sheep are collected, and corresponding three-dimensional sheep face models are generated using three-dimensional reconstruction technology, further obtaining three-dimensional sheep face images from three different perspectives. Additionally, this study developed a sheep face orientation recognition algorithm called the sheep face orientation recognition algorithm (SFORA). The SFORA incorporates the ECA mechanism to further enhance recognition performance. Ultimately, the SFORA has a model size of only 5.3 MB, with accuracy and F1 score reaching 99.6% and 99.5%, respectively. During the recognition task, the SFORA is first used for sheep face orientation recognition, followed by matching the recognition image with the corresponding three-dimensional sheep face image based on the established SuperGlue feature-matching algorithm, ultimately outputting the recognition result. Experimental results indicate that when the confidence threshold is set to 0.4, SuperGlue achieves the best matching performance, with matching accuracies for the front, left, and right faces reaching 96.0%, 94.2%, and 96.3%, respectively. This study enriches the theoretical research on sheep face recognition technology and provides technical support.