Search

To enhance the accuracy of fine-grained classification of shark populations and address issues such as image interference, insufficient extraction of local key features, and lack of semantic correlation between channels, a DM-BCNN model for fine-grained classification of shark populations based on an improved Bilinear Convolutional Neural Network (B-CNN) is proposed. First, deformable convolution is introduced to replace the feature extraction part of the original model with a DRAM_ResNet network structure, enhancing the model's ability to detect complex and irregular shapes and local structures. Then, the NAM attention mechanism is employed to strengthen the model's ability to identify and extract key features. Finally, a Mutual Channel Loss function is introduced to enhance the semantic correlation between different channels of shark images, allowing the model to capture information from various aspects of the images more comprehensively. The results show that the improved DM-BCNN model achieved a Top-1 accuracy of 96. 12%, representing a 2.51 percentage point improvement over the original model. The study demonstrates that the proposed improved model outperforms the original model in fine-grained image classification, making it more effective for fine-grained classification and identification of shark populations. [ABSTRACT FROM AUTHOR]

Fine-grained few-shot image classification is a popular research area in deep learning. The main goal is to identify subcategories within a broader category using a limited number of samples. The challenge stems from the high intra-class variability and low inter-class variability of fine-grained images, which often hamper classification performance. To overcome this, we propose a fine-grained few-shot image classification algorithm based on bidirectional feature reconstruction. This algorithm introduces a Mixed Residual Attention Block (MRA Block), combining channel attention and window-based self-attention to capture local details in images. Additionally, the Dual Reconstruction Feature Fusion (DRFF) module is designed to enhance the model's adaptability to both inter-class and intra-class variations by integrating features of different scales across layers. Cosine similarity networks are employed for similarity measurement, enabling precise predictions. The experiments demonstrate that the proposed method achieves classification accuracies of 96.99%, 98.53%, and 89.78% on the CUB-200-2011, Stanford Cars, and Stanford Dogs datasets, respectively, confirming the method's efficacy in fine-grained classification tasks. [ABSTRACT FROM AUTHOR]

3D object detection is a critical task in the fields of virtual reality and autonomous driving. Given that each sensor has its own strengths and limitations, multi-sensor-based 3D object detection has gained popularity. However, most existing methods extract high-level image semantic features and fuse them with point cloud features, focusing solely on consistent information from both sensors while ignoring their complementary information. In this paper, we present a novel two-stage multi-sensor deep neural network, called the adaptive learning point cloud and image diversity feature fusion network (APIDFF-Net), for 3D object detection. Our approach employs the fine-grained image information to complement the point cloud information by combining low-level image features with high-level point cloud features. Specifically, we design a shallow image feature extraction module to learn fine-grained information from images, instead of relying on deep layer features with coarse-grained information. Furthermore, we design a diversity feature fusion (DFF) module that transforms low-level image features into point-wise image features and explores their complementary features through an attention mechanism, ensuring an effective combination of fine-grained image features and point cloud features. Experiments on the KITTI benchmark show that the proposed method outperforms state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Current single-model methods for fine-grained image classification suffer from insufficient generalisation ability, while multi-model fusion methods suffer from weight curing. The study suggests and experimentally tests a dynamic weight multi-model fusion strategy for transfer learning-based fine-grained picture classification. The results of the experiment showed that the suggested fusion model enhanced recognition accuracy by 1.33%, 1.19%, and 0.83% compared to the single model on the medical dataset and 3.25%, 1.34%, and 7.28% on the agronomy dataset, respectively. Furthermore, when compared to the comparison method, the models under the proposed method of the study improved recognition accuracy by 0.18%, 0.61%, 0.43%, and 0.43% on the medical dataset, and the experimental time consumed was 3.25 minutes less than that of the sum-of-maximum-probabilities method; however, the fusion models of the proposed method of the study had higher recognition accuracy than that of the comparison met Overall, the proposed dynamic weight multi-model fusion method for fine-grained image classification using migration learning has better performance and generalisation ability, which can improve performance while reducing time cost, and has higher application value for the actual fine-grained image classification task.

Abstract 3D object detection is a critical task in the fields of virtual reality and autonomous driving. Given that each sensor has its own strengths and limitations, multi-sensor-based 3D object detection has gained popularity. However, most existing methods extract high-level image semantic features and fuse them with point cloud features, focusing solely on consistent information from both sensors while ignoring their complementary information. In this paper, we present a novel two-stage multi-sensor deep neural network, called the adaptive learning point cloud and image diversity feature fusion network (APIDFF-Net), for 3D object detection. Our approach employs the fine-grained image information to complement the point cloud information by combining low-level image features with high-level point cloud features. Specifically, we design a shallow image feature extraction module to learn fine-grained information from images, instead of relying on deep layer features with coarse-grained information. Furthermore, we design a diversity feature fusion (DFF) module that transforms low-level image features into point-wise image features and explores their complementary features through an attention mechanism, ensuring an effective combination of fine-grained image features and point cloud features. Experiments on the KITTI benchmark show that the proposed method outperforms state-of-the-art methods.

The fine-grained visual recognition is to classify several sub-categories affiliated to the same basic-level category, which is highly challenging because the same sub-category with large variance and different sub-categories with small variance. Previously approaches generally localize the targets or parts first, then determine which sub-category the image is attached to. They depend on target or part annotations, which are labor-intensive and a barrier to moving towards practical use. Other methods indirectly extract recognizable areas from the high-level feature maps, ignoring the spatial relationships between the target and its parts, which may cause inaccurate recognition. In this paper, we propose a weakly supervised spatial group attention network (WSSGA-Net) for fine-grained bird recognition. According to the spatial relationships between the target and its parts, we embed the spatial group attention (SGA) module into the WSSGA-Net to highlight the correct semantic feature regions by establishing a semantic feature space enhancement mechanism. In addition, we apply moment exchange (MoEx) to generate new feature maps by exchanging two input image feature moments for data augmentation. Comprehensive experiments indicate that our approach significantly has a better performance than the state-of-the-art approaches on the standard bird image datasets Bird-65, CUB200-2011 and fine-grained dataset Stanford Cars. [ABSTRACT FROM AUTHOR]

Fine-grained image classification is a challenging research topic because of the high degree of similarity among categories and the high degree of dissimilarity for a specific category caused by different poses and scales. A cultural heritage image is one of the fine-grained images because each image has the same similarity in most cases. Using the classification technique, distinguishing cultural heritage architecture may be difficult. This study proposes a cultural heritage content retrieval method using adaptive deep learning for fine-grained image retrieval. The key contribution of this research was the creation of a retrieval model that could handle incremental streams of new categories while maintaining its past performance in old categories and not losing the old categorization of a cultural heritage image. The goal of the proposed method is to perform a retrieval task for classes. Incremental learning for new classes was conducted to reduce the re-training process. In this step, the original class is not necessary for re-training which we call an adaptive deep learning technique. Cultural heritage in the case of Thai archaeological site architecture was retrieved through machine learning and image processing. We analyze the experimental results of incremental learning for fine-grained images with images of Thai archaeological site architecture from world heritage provinces in Thailand, which have a similar architecture. Using a fine-grained image retrieval technique for this group of cultural heritage images in a database can solve the problem of a high degree of similarity among categories and a high degree of dissimilarity for a specific category. The proposed method for retrieving the correct image from a database can deliver an average accuracy of 85 percent. Adaptive deep learning for fine-grained image retrieval was used to retrieve cultural heritage content, and it outperformed state-of-theart methods in fine-grained image retrieval. [ABSTRACT FROM AUTHOR]

In order to solve the problem of low recognition accuracy in fine-grained ship images with a single feature, a ship target recognition method based on the fusion of recurrent attention convolutional neural network (RA-CNN) and multi- feature regions is proposed. This method introduces the scale-dependent pooling (SDP) algorithm in the VGG-19 network to solve the problem of excessive pooling of small targets in the network, and improves the recognition performance of small ships. The attention proposal network (APN) introduces joint clustering algorithm to form multiple independent feature regions, so that the whole model can make full use of global information and improve the accuracy of ship recognition. At the same time, a feature region optimization method is designed to reduce the overlap rate of multiple feature region and solve the problem of overfitting. By defining a new loss function to crosstrain the VGG-19 and APN, the convergence is accelerated. The proposed method is tested by using the open optoelectronic ship database, the recognition accuracy is up to 90.2%. Both the recognition rate and the robustness of the model are greatly improved compared with the single feature. [ABSTRACT FROM AUTHOR]

Hand bone age, as the biological age of humans, can accurately reflect the development level and maturity of individuals. Bone age assessment results of adolescents can provide a theoretical basis for their growth and development and height prediction. In this study, a deep convolutional neural network (CNN) model based on fine-grained image classification is proposed, using a hand bone image dataset provided by the Radiological Society of North America (RSNA) as the research object. This model can automatically locate informative regions and extract local features in the process of hand bone image recognition, and then, the extracted local features are combined with global features of a complete image for bone age classification. This method can achieve end-to-end bone age assessment without any image annotation information (except bone age tags), improving the speed and accuracy of bone age assessment. Experimental results show that the proposed method achieves 66.38% and 68.63% recognition accuracy of males and females on the RSNA dataset, and the mean absolute errors are 3.71 ± 7.55 and 3.81 ± 7.74 months for males and females, respectively. The test time for each image is approximately 35 ms. This method achieves good performance and outperforms existing methods in bone age assessment based on weakly supervised fine-grained image classification.

In the existing metric learning methods, the tuple-based loss training speed is slow and the proxy-based loss don’t consider the fine-grained semantic relationship between the data. In response to these problems, the paper combined the advantages of the two and proposed a DRPLoss function for fine-grained images. This paper used an inception network with a BN layer as the embedding network, learnt the correlation between data through gradient interaction in the metric space, and used temperature scaling to adjust the DRPLoss to supervise and train the embedding vector. Finally, this paper verified the effectiveness of different embedding dimensions DRPLoss on CUB-200-2011 and Car-196 datasets. The experiment improves the evaluation index of recall@1 by 2% and 6.4% respectively. Compared with the tuple-based loss and proxy-based loss, the experimental results show that DRPLoss is faster in training and has a significant improvement in the performance of fine-grained image retrieval. [ABSTRACT FROM AUTHOR]

Real-time dynamic monitoring of orchard grape leaf diseases can greatly improve the efficiency of disease control and is of great significance to the healthy and stable development of the grape industry. Traditional manual disease-monitoring methods are inefficient, labor-intensive, and ineffective. Therefore, an efficient method is urgently needed for real-time dynamic monitoring of orchard grape diseases. The classical deep learning network can achieve high accuracy in recognizing grape leaf diseases; however, the large amount of model parameters requires huge computing resources, and it is difficult to deploy to actual application scenarios. To solve the above problems, a cross-channel interactive attention mechanism-based lightweight model (ECA-SNet) is proposed. First, based on 6,867 collected images of five common leaf diseases of measles, black rot, downy mildew, leaf blight, powdery mildew, and healthy leaves, image augmentation techniques are used to construct the training, validation, and test set. Then, with ShuffleNet-v2 as the backbone, an efficient channel attention strategy is introduced to strengthen the ability of the model for extracting fine-grained lesion features. Ultimately, the efficient lightweight model ECA-SNet is obtained by further simplifying the network layer structure. The model parameters amount of ECA-SNet 0.5× is only 24.6% of ShuffleNet-v2 1.0×, but the recognition accuracy is increased by 3.66 percentage points to 98.86%, and FLOPs are only 37.4 M, which means the performance is significantly better than other commonly used lightweight methods. Although the similarity of fine-grained features of different diseases image is relatively high, the average F1-score of the proposed lightweight model can still reach 0.988, which means the model has strong stability and anti-interference ability. The results show that the lightweight attention mechanism model proposed in this paper can efficiently use image fine-grained information to diagnose orchard grape leaf diseases at a low computing cost. [ABSTRACT FROM AUTHOR]

Real-time dynamic monitoring of orchard grape leaf diseases can greatly improve the efficiency of disease control and is of great significance to the healthy and stable development of the grape industry. Traditional manual disease-monitoring methods are inefficient, labor-intensive, and ineffective. Therefore, an efficient method is urgently needed for real-time dynamic monitoring of orchard grape diseases. The classical deep learning network can achieve high accuracy in recognizing grape leaf diseases; however, the large amount of model parameters requires huge computing resources, and it is difficult to deploy to actual application scenarios. To solve the above problems, a cross-channel interactive attention mechanism-based lightweight model (ECA-SNet) is proposed. First, based on 6,867 collected images of five common leaf diseases of measles, black rot, downy mildew, leaf blight, powdery mildew, and healthy leaves, image augmentation techniques are used to construct the training, validation, and test set. Then, with ShuffleNet-v2 as the backbone, an efficient channel attention strategy is introduced to strengthen the ability of the model for extracting fine-grained lesion features. Ultimately, the efficient lightweight model ECA-SNet is obtained by further simplifying the network layer structure. The model parameters amount of ECA-SNet 0.5× is only 24.6% of ShuffleNet-v2 1.0×, but the recognition accuracy is increased by 3.66 percentage points to 98.86%, and FLOPs are only 37.4 M, which means the performance is significantly better than other commonly used lightweight methods. Although the similarity of fine-grained features of different diseases image is relatively high, the average F1-score of the proposed lightweight model can still reach 0.988, which means the model has strong stability and anti-interference ability. The results show that the lightweight attention mechanism model proposed in this paper can efficiently use image fine-grained information to diagnose orchard grape leaf diseases at a low computing cost.

The fine-grained image classification method does not have a very good training effect when performing random weight training for neural networks. For the fine-grained classification under the condition of convolutional neural network, the main reason is that its upgrading method can better improve the stability of convolutional neural network. Fine-grained classification through bilinear convolutional neural networks is to effectively merge several convolutional neural networks, so as to obtain a high-resolution advantage to improve the method of fine-grained classification. With the development of society and the improvement of urban construction, the artificial reconstruction of rivers in the city will strengthen the connectivity and smoothness of the reconstructed river to a certain extent, but it will also affect the self-healing of the river. At present, the treatment of river pollution is a major aspect of the country's treatment of environmental pollution. Obtain the transformation and restoration of the ecological environment around the river in the city through data exchange and other methods. Establishing a scientific and complete plant community with a naturalized revetment system can reduce the impact of river water on the riverbank. At the same time, these naturalized revetment plants can also play a role in flood prevention and discharge, and to a certain extent, it can affect the polluted water quality and give purification. While carrying out the ecological environment protection of the river, it also pays attention to the design and planning of the river landscape, so that the river can withstand the impact of flowing water while having a beautiful landscape design, so as to realize the ecological civilization construction of the river landscape. [ABSTRACT FROM AUTHOR]

Although the performance of object detectors is constantly improving based on benchmark assessments, how to carry out their industrial deployment remains a challenging task. Considering that fine-grained characteristic is widely existed in industrial object images, we proposed a single-shot fine-grained object detector, and applied it to coal–gangue images in coal preparation plant. Firstly, we employed grouped convolution to build a new convolutional neural network as the backbone of the detector. Secondly, a feature fusion module, which was designed based on spatial attention, was employed to optimize the features extracted by a feature pyramid network. Finally, a feature separation module, which was designed based on channel attention, was employed to alleviate the conflict in detection between the classification task and the localization task. Moreover, the effect of each component was fully demonstrated by ablation study, visualization analysis and hypothesis testing. Compared with other classical detectors, the proposed detector's AP iou=0.5 has increased by an average of approximately 10% in coal–gangue dataset. • A new method is proposed for detecting fine-grained objects in industry. • The large differences within a class can be solved by grouped convolution. • Spatial-attention is applied to optimize the feature fusion. • Channel-attention is applied to alleviate the feature conflict in object detection. • Designing a detector based on coal–gangue datasets is superior to the existing classical detectors. [ABSTRACT FROM AUTHOR]