Similar Journals
|
Signal, Image and Video Processing
Journal Prestige (SJR): 0.485  Citation Impact (citeScore): 2 Number of Followers: 11  Hybrid journal (It can contain Open Access articles)ISSN (Print) 1863-1711 - ISSN (Online) 1863-1703 Published by Springer-Verlag  [2468 journals]
|
- SW-Net: anchor-free ship detection based on spatial feature enhancement
and weight-guided fusion-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Synthetic aperture radar (SAR) images are widely used for maritime surveillance due to their all-weather imaging capability and day–night visibility. However, the sparsity of offshore scene targets and the interference of land facilities in inshore scenes increase the difficulty of SAR ship detection, and the anchor-based detection algorithms require a large amount of computational resources. Therefore, this paper proposes an anchor-free detection method for SAR ship detection based on spatial feature enhancement and weight-guided fusion, called SW-Net. First, a spatial feature enhancement module is constructed to reduce the information loss caused by a sudden decrease in the number of feature channels by enhancing the spatial structural information of the features. Additionally, to solve the problem of blurred target boundaries after fusing features of different scales, a weight-guided fusion module is designed to use high-level features to generate weight vectors to guide the fusion of low-level features and generate more powerful semantic information. Finally, the complete intersection over union loss function is utilized to optimize the predicted boxes, to increase their quality. We performed experiments on the SSDD and HRSID public datasets to evaluate SW-Net’s performance. The results of our experiments show that SW-Net consistently surpasses existing methods as a matter of detection accuracy, demonstrating the efficacy of our suggested approach.
PubDate: 2023-12-07
-
- A lightweight robust image hash based on random tensors and angle features
for IoT devices-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Image hashing can be extensively used in image forensics, and the lightweight image hash suitable for IoT smart devices plays an important role of integrity verification for the image obtained by these devices; however, the robust image hashing used in smart devices with resource-constrained is rarely discussed. Therefore, a lightweight perceptual robust image hashing scheme based on random tensor and angle features is introduced in this paper. Specifically, the global features are obtained through quantization of DCT coefficients generated from expanding of two skillfully designed three-order tensors; in order to obtain image local features, the energies of some non-overlapping image blocks are first computed, and then, local features are achieved through calculating the angle features. At last, the global and local features are transformed into the corresponding hash. Large quantities of experiments on four datasets are implemented to justify the effectiveness of the proposed method. Some comparisons on length of hashing, TPR, FPR, ROC curve, AUC value and collision probability show that the suggested proposal achieves moderate length, higher detection accuracy and better balance between robustness and discrimination than the state-of-the-art algorithms, and the tests in smart phone-based data collecting system show that the proposed technique is feasible, and it has potential application value for image verification in IoT devices.
PubDate: 2023-12-07
-
- KinD-LCE: curve estimation and Retinex Fusion on low-light image
-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Low-light images often suffer from noise and color distortion. Object detection, semantic segmentation, instance segmentation, and other tasks are challenging when working with low-light images because of image noise and chromatic aberration. We also found that the conventional Retinex theory loses information in adjusting the image for low-light tasks. In response to the aforementioned problem, this paper proposes an algorithm for low illumination enhancement. The proposed method, KinD-LCE, uses a light curve estimation module to enhance the illumination map in the Retinex decomposed image, improving the overall image brightness. An illumination map and reflection map fusion module were also proposed to restore the image details and reduce detail loss. Additionally, a TV(total variation) loss function was applied to eliminate noise. Our method was trained on the GladNet dataset, known for its diverse collection of low-light images, tested against the Low-Light dataset, and evaluated using the ExDark dataset for downstream tasks, demonstrating competitive performance with a PSNR of 19.7216 and SSIM of 0.8213.
PubDate: 2023-12-06
-
- A histogram equalization model for color image contrast enhancement
-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract The main aim of this paper is to develop a histogram equalization algorithm for color image contrast enhancement. Our idea is to propose a variational approach containing an energy functional to determine local transformations in the lightness (L) and chroma (C) channels of the CIE LCH color space such that the histograms in these two channels can be redistributed locally. In order to minimize the differences among the local transformation at the nearby pixel locations in each channel, the spatial regularization of the transformation is incorporated in the functional for the equalization process. The existence and uniqueness of the minimizer of the variational model can be shown. Experimental results are reported to show that the performance of the proposed models is competitive with the other compared methods for several testing images.
PubDate: 2023-12-05
-
- Separable feature complementary network with branch-wise and multi-scale
spatial attention for lightweight image super-resolution-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Recent advances in single image super-resolution (SISR) have shown promising results, but networks with optimal performance tend to have heavy computation, making them unsuitable for edge devices. How to achieve better results with fewer parameters is still a problem that requires further research. To overcome this issue, we propose a separable feature complementary network using branch-wise attention and multi-scale spatial attention (SFCN-BMSA). The network contains a feature complementary module, which utilizes a limited number of small-sized convolution kernels to combine long-range features from different positions on the feature map and utilizes them to enhance image reconstruction. In addition, we design a feature fusion module with branch-wise attention, which can fuse the features of different branches according to the importance of each branch. Finally, we also design a multi-scale spatial attention module, which utilizes three dilated convolutions with the size of 5 \(\times \) 5 to calculate attention from different spatial scales and combines them to obtain more refined attention while utilizing a larger receptive field. Experiments show that the proposed neural network achieves better reconstruction results with lower parameters.
PubDate: 2023-12-05
-
- Decadal forest cover change analysis of the tropical forest of
Tadoba-Andhari, India-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Deforestation is a major concern for preserving the biodiversity of the entire globe. During the last few years, machine learning and deep learning methods have been employed for mapping deforestation. There is still scope for ample improvement in these methods as they are prone to errors and can give inaccurate results because of over or under-segmentation. This paper uses deep convolutional neural network-based semantic segmentation to process multispectral satellite images to monitor forest cover changes in Tadoba-Andhari National Park during the period 2000–2022. The proposed approach uses the U-Net architecture with extended inputs which gives more accuracy as compared to U-Net with only image input. Landsat images along with vegetation indices have been used as training data. The proposed method requires less time to train the model and is also cost-efficient in terms of computing requirements. The performance of the proposed method was compared with state-of-the-art methods where the proposed method outperformed the other models with an F1-score of 0.90 and an accuracy of 84.83%. When compared with U-Net trained with Landsat images only, it was observed that the U-Net model trained with extended input was able to achieve better results.
PubDate: 2023-12-01
-
- An adaptive guidance fusion network for RGB-D salient object detection
-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract RGB-D salient object detection (RGB-D SOD) has currently attracted much attention for its prospect of broad application. On the basis of the “encoder-decoder” paradigm of the fully convolutional network (FCN), many FCN-based strategies have emerged and achieved huge progress, but underestimated the potential of level-specific characteristics of multi-modal features. In this paper, we propose the adaptive guided fusion network (AGFNet) to further mine the potential information between the depth image and the RGB image, and design an adaptive fusion and coarse-to-fine decoding strategy to achieve high-precision detection of salient objects. Specifically, we first use a two-stream encoder to extract the multi-level features of the RGB image and depth image but refrain from the previous practice of using depth features for each layer. Second, a simple but effective way named multi-modal selective fusion strategy is designed to fuse the multi-level features. Third, for enhancement of contextual information of each level adaptively, an adaptive cross fusion module (ACFM) fuses the features at all levels and outputs a coarse saliency map. Finally, a guided attention refinement module (GARM) utilizes the coarse saliency map to guide the final features from ACFM to realize the enhancement and obtain a refined saliency map. Our method is compared with other state-of-the-art RGB-SOD methods through extensive experiments, and the results demonstrate the superiority of our proposed AGFNet. The source code of this project is available at https://github.com/HaodongSun809/my_AGFNet.git.
PubDate: 2023-11-30
-
- Cross-propagation parallel network for reflection image inpainting
-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract The development of deep learning has led to great advances in image reflection removal techniques. Most existing image reflection removal methods can produce reasonable results but still suffer from missing structure and blurry textures. The main reasons are the following facts: (1) Separation-based methods struggle to distinguish fine structures due to the similarity between reflection and background layers. (2) U-Net may cause the reintroduction of reflection features. To tackle above issues, a novel Cross-Propagation Parallel Network (CPPN) is proposed for reflection image inpainting. Firstly, the correct information from the reflection-free regions of the blended image is used to infer information from the reflective regions based on image inpainting mechanisms to improve the accuracy of structural features. Secondly, a cross-propagation parallel network is designed to further constrain structure and detail for accurate feature representation and propagation. Experimental results on several public datasets show that the proposed method can produce higher-quality results than state-of-the-art methods.
PubDate: 2023-11-30
-
- Performance analysis of MTS on the VVC encoder
-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Versatile video coding standard known as H.266/VVC has been adopted by the Joint Video Experts Team in July 2020 Bross et al. (ITU-T and ISO/IEC JVET-S2001, 2020). It is the latest video coding standard with advanced tools to enhance coding efficiency. Notably, H.266/VVC introduces new techniques, including Multiple Transform Selection (MTS). In this paper, detailed descriptions of the new transform coding development in VVC standard are presented. The study also highlights the impact of the MTS on the VVC encoder performance through experimental research using VVC reference software VTM-14 for random access configuration. The experimental results show an increase of 8.35% in terms of time execution with an improvement of 1% in the quality of the reconstructed video along with a reduction of 1% in bitrate for a quantification parameter QP equal to 32. For a QP equal to 22, there is an increase of 15.8% in terms of time execution with an improvement of 0.056% in the quality of the reconstructed video along with a reduction of 0.445% in bitrate.
PubDate: 2023-11-29
-
- Feature channel interaction long-tailed image classification model based
on dual attention-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract In the real world, the data distribution often presents a long tail distribution, and the imbalance of data will lead to the model learning bias to the head class. To address the influence of long tail distribution on image classification, this paper proposes a feature channel interactive long tail image classification model based on dual attention. Firstly, the dual attention module is used to capture the autocorrelation and spatial dimension information of the feature map, and the enhanced image is obtained by transformation and class activation map. After that, image preprocessing is performed on the enhanced data set to reduce the over-fitting of the model to the head, and the features that are more conducive to tail classification are obtained through learning. Finally, by interacting with the local channels adjacent to the features, the correlation between the channels is extracted to obtain more robust features. The method achieves good performance on CIFAR10-LT, CIFAR100-LT and ImageNet datasets, which proves the effectiveness of the model.
PubDate: 2023-11-29
-
- Influence of sorting measures on similar segment grouping based denoising
algorithms-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Denoising is a fundamental problem in digital image processing and computer vision. Many of the denoising algorithms use segment-based methods. Various methods for nonlocal similar segments, namely identification, sorting and grouping, are used for efficient denoising. The basis of these denoising algorithms is the existence of nonlocal similarity and redundancy between the pixels. The sorting measures, segment size and the number of segments influence the performance in this method. This algorithm evaluation paper presents the similarity measures used in various similar segment grouping-based denoising methods and the effect of various norms, segment size and segment groups on the performance these algorithms under different noise levels.
PubDate: 2023-11-28
-
- A colour image segmentation method and its application to medical images
-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract In this paper, we propose a segmentation model using an anisotropic multi-well potential-based nonlinear transient PDE for colour images. A channel-wise greyscale classification approach is devised for colour image segmentation. The time evolution of the PDE model is carried out by the implicit–explicit convexity splitting approach. Further, we consider the fractional version of the time-discretised model by replacing the Laplacian with its fractional counterpart. The spatial terms are approximated by the Fourier basis under the pseudo-spectral method. The convergence and the stability of the numerical scheme are elaborated. Both models (fractional and non-fractional) are tested on some synthetic images and few real-world standard test images. The results on synthetic images are compared with those from the literature using Dice similarity index, Jaccard similarity index and BF score. Later the method is successfully applied on several medical images to classify the same.
PubDate: 2023-11-28
-
- Hyperspectral image synthesis from sparse RGB data: a comparative study
combining linear regression, multilayer perceptron, and clustering-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract The problem of synthesizing hyperspectral images from RGB images is ill posed, with potentially infinite solutions, as it involves estimating data in a high-dimensional space, associated with hyperspectral bands, from limited information in a three-dimensional RGB space. However, under certain conditions related to lighting and physical properties of natural scenes, a feasible solution can be found. This study evaluates four methods for estimating hyperspectral data from RGB images: ridge linear regression, Minibatch K-means followed by linear regression, a multilayer perceptron (MLP) neural network, and Minibatch K-means combined with an MLP neural network. The results of each method are compared with each other and with the NTIRE 2020 Challenge. The comparison was performed using the mean absolute relative error (MARE) and execution time. The MLP method attained the lowest MARE (0.072) but with the longest execution time (220 s). Ridge regression attained the shortest execution time (0.47 s) at the cost of a higher MARE (0.089). The best trade-off was obtained by combining Minibatch K-means clustering with MLP, which reduced the execution time by 16 times (13.8 s) with a slightly higher MARE (0.075) compared to MLP alone. We have also confirmed that, for the case of natural scenes, points representing pixels that are close to each other in the RGB space are also close to each other in the hyperspectral space.
PubDate: 2023-11-26
-
- A lightweight and robust block cipher algorithm for real-time applications
-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Ensuring the protection of big data during transmission or storage represents a highly significant concern for both enterprises and individual users. Cryptography is employed to convert data into a format that remains comprehensible solely to its intended recipients. The widely adopted Advanced Encryption Standard (AES) stands as one of the most robust encryption algorithms available. Nevertheless, its applicability to real-time applications is hindered by speed-related challenges and computational intricacies. This paper presents an innovative real-time lightweight symmetric block cipher algorithm based on a novel transformation named Mix-data, a low-cost random permutation, an S-box substitution, and an XOR operation with a robust key. The proposed cryptosystem is implemented on the Altera Cyclone III EP3C80F780C8 hardware platform in conjunction with a Raspberry Pi 4 software platform. The system implementation on the Altera Cyclone III FPGA utilizes only 4.851 of the total logic elements, consumes 14,974 mW of total power dissipation and achieves an impressive throughput of 15.81 Gbit/s. Rigorous security evaluations have been conducted to demonstrate its robustness against statistical and differential attacks. The evaluation results confirm the effectiveness, speed, and high level of security of the proposed algorithm.
PubDate: 2023-11-26
-
- End-to-end deep learning pipeline for scalable, deployable object
detection engine in the traffic system-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract A wide variety of vehicles are moving on roads, and these vehicles are classified and detected by intelligent traffic systems using object detection. Various traditional methods are inefficient in producing good accuracy for detecting vehicles in traffic due to a lack of object discrimination capability. Object detection algorithms typically rely on deep convolutional neural networks, which require the host device with high computing capabilities, significantly limiting the applications of object detection algorithms for edge devices with limited computing capabilities. There is a need for an object detector to address the problem of detecting objects in traffic. This paper proposes a novel framework consisting of a deep learning model with the training-to-inference pipeline for object detection on images acquired from Indian city streets. The deep learning model has been used to streamline the conversion of the trained model to an optimized Triton-compatible (TensorRT) model. The Triton Inference Server model is evaluated on the Indian driving dataset (IDD) with NVIDIA Jetson AGX Xavier edge device. The model can easily deploy on traffic data collected from any client devices located remotely. The experimental results show that the Triton Inference outperformed the existing techniques. The proposed framework can be used to prevent traffic congestion. The proposed method achieved a precision of 93.8%, recall of 76.5%, and MAP of 80.5%, respectively, on IDD images with a resolution of 1920 × 1080.
PubDate: 2023-11-25
-
- Image denoising based on the fractional-order total variation and the
minimax-concave-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract The total variation model has attracted considerable attention for its good balance of noise reduction and edge maintenance, but it produces blocky effects. In this paper, a novel model for noise reduction and staircase effects elimination was proposed for images polluted by the additive white Gaussian noise, which is based on fractional-order differentiation. The new non-convex regularization term can express as the minimax-concave penalty of the fractional-order total variation (FOTV) term. The FOTV term can suppress staircase effects while preserving small-scale edges and textures information well. The non-convex regularizer can estimate the edge more accurately than the convex regularizer. We set the non-convexity parameter and the regularization parameter in the appropriate range to maintain the convexity of the proposed objective function. To effectively solve the new model, we use the alternating direction method of multipliers to minimize the objective function. Experimental results illustrate that the new model performs better than other models and yields clearer denoised images.
PubDate: 2023-11-25
-
- Epileptic seizure detection using scalogram-based hybrid CNN model on EEG
signals-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Epilepsy is one of the most usual neurological diseases characterized by abnormal brain activity, resulting in seizures or strange behavior, sensations, and, in some cases, loss of consciousness. It is a persistent, non-communicable brain condition that can affect anyone at any age, nearly 50 million people globally, with about 80% of sufferers living in low- and middle-income countries. Electroencephalography (EEG) signals are largely used in epilepsy research to examine brain activity during seizures. The extraction of features and selection from EEG signals plays a major role in epileptic seizure detection. In traditional machine learning techniques, the hard-core feature extraction needs domain expertise, and this can be eliminated by deep learning. The benefits of deep learning techniques are they try to learn high-level features from the input signals in an incremental method. To meet the requirements of complicated feature engineering, deep learning techniques have received greater attention than conventional methods. A hybrid seizure detection-convolutional neural network and vector machine (SD-CNN and SVM) model is proposed for epileptic seizure detection with EEG signals. Transformation of signal to image is performed using continuous wavelet transform technique to generate scaleogram images and also SD-CNN works as a learnable feature extractor from the generated images and SVM works as a binary classifier. The experimental results extracted 94% with high quality of scaleogram images using hybrid SD-CNN and SVM model and removed the noise levels and time–frequency data from EEG signals.
PubDate: 2023-11-24
-
- A key-points-assisted network with transfer learning for precision human
action recognition in still images-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Still image-based human action recognition is a highly sought-after but challenging field in computer vision, and such challenge mainly stems from the lack of information in single images. Therefore, efficient extraction of visual appearance features and other valuable information of images is crucial for action recognition. To this purpose, on the one hand, we use a convolutional neural network (CNN) classifier based on EfficientNetV2-S network as the main pathway for extracting appearance features from images and classification. To make the CNN classifier focus on important spatial features, we propose the residual spatial attention module (RSAM) and incorporate it into the CNN classifier. In addition, we leverage transfer learning to enhance the training speed and recognition precision of the CNN classifier. On the other hand, we utilize the OpenPose algorithm to extract the coordinates of human key-points in the auxiliary pathway and perform information extraction and classification on the obtained key-points with a self-made network. Finally, we use one-dimensional convolution to merge the results of these two classifications. One-dimensional convolution can automatically learn the weights of these two results and merge them based on their importance. Experimental results on three challenging datasets, namely Stanford 40 Actions, People Play Music Instrument (PPMI) and MPII Human Pose datasets, illustrate the superiority of the proposed method.
PubDate: 2023-11-22
-
- Weed detection in agricultural fields via automatic graph cut segmentation
with Mobile Net classification model-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract Agriculture is heavily affected by weeds due to their random appearances in fields, competition for water, nutrients, and sunlight, and, if not controlled effectively, negative impact on crop yields. In general, there are many prevention strategies, but they are expensive and time consuming; moreover, labor costs have increased substantially. To overcome these challenges, a novel AGS-MNFELM model has been proposed for weed detection in agricultural fields. Initially, the gathered images are pre-processed using bilateral filter for noise removal and CLAHE for enhancing the image quality. The pre-processed images are taken as an input for automatic graph cut segmentation (AGS) model for segmenting regions with bounding box using the RCNN rather than manual initialization, hence eliminating the need for manual interpretation. The Mobile Net model is used to acquired rich feature representations for a variety of images, and the retrieved features FELM (Fuzzy Extreme Learning Machine Model) classifier is used to classify four weed types of maize and soyabean: cocklebur, redroot pigweed, foxtail, and giant ragweed. The proposed AGS-MNFELM model has been evaluated in terms of its sensitivity, accuracy, specificity, and F1 score. The experimental result reveals that the proposed AGS-MNFELM model attains the overall accuracy of 98.63%. The proposed deep learning-based MobileNet improves the overall accuracy range of 7.91%, 4.15%, 3.44% and 5.88% better than traditional LeNet, AlexNet, DenseNet and ResNet, respectively.
PubDate: 2023-11-21
-
- Freshness uniformity measurement network based on multi-layer feature
fusion and histogram layer-
Free pre-print version: Loading...Rate this result: What is this?Please help us test our new pre-print finding feature by giving the pre-print link a rating.
A 5 star rating indicates the linked pre-print has the exact same content as the published article.
Abstract: Abstract The arrangement of products on supermarket freshness shelves exhibits a certain pattern and displays distinct texture characteristics. In recent years, many studies have applied texture extraction algorithms in deep learning, such as the Histogram Layer Residual Network (HistNet). However, this algorithm still has obvious disadvantages, such as neglecting the optimal representation of multi-scale texture features and lacking feature selection during extraction. To address these issues, this paper introduces a novel texture classification network—Multi-Scale Feature Histogram Network (MFHisNet). First, we design a Multi-Scale Feature Fusion Module (MF-Block) to achieve a multi-level representation of texture information. Then, we utilize an attention module (CBAM) to weight crucial information and suppress background interference for deeper level texture features. Experimental results demonstrate that the model achieves accuracies of 82.12 ±2.04 \(\%\) , 73.13±1.10 \(\%\) , and 83.46±0.62 \(\%\) on the GTOS-mobile, DTD, and MINC-2500 datasets, respectively. Furthermore, based on the proposed model, we propose a measurement method that uses cosine similarity to measure the uniformity of freshness placement, and the effectiveness of this method was verified on the dataset we collected.
PubDate: 2023-11-19
-
