Image Classification Performance Research Articles

Convolutional neural network classification of beam profiles from silicon photonics gratings

Convolutional neural network (CNN) models consist of CNN block(s), and dense neural network (DNN) block(s) are used to perform image classification on beam profiles in light beams coupled out from silicon photonics (SiPh) mixed-pitch gratings. The beam profiles are first simulated and segregated into three categories based on their corresponding height above the SiPh gratings. With one CNN block, one DNN block, and 128 nodes in the DNN block, classification accuracy of 98.68% is achieved when classifying 454 beam profile images to their corresponding categories. Expanding the number of CNN blocks, DNN blocks, and nodes, 64 CNN models are constructed, trained, and evaluated. Out of the 64 CNN models, 52 of them achieved classification accuracy of >95%.

Impact the Classes’ Number on the Convolutional Neural Networks Performance for Image Classification

Deep learning was developed as a realistic artificial intelligence technique that takes in numerous layers of information and produces the best results in various classes. Deep learning has demonstrated excellent performance in several areas, particularly picture grouping, division, and recognition. The convolutional neural network (CNN) is one of the algorithms that relies on deep learning in its work. It has proven its effectiveness in classifying images with high efficiency in medical images and their diagnoses, face recognition, and other different fields. In this paper, the focus was on images to alert new researchers to their effects on the performance of CNN in terms of the number of classes that existed within the database, in addition to the impact of incorrect classification of images by the source on the classification result and the necessity of adopting reliable and correct sources of data to avoid inaccurate results. A group of face images has been used, and three experiments on them were conducted using all existing classes with reduction. The results showed a significant improvement in the performance of the algorithm whenever the number of classes was reduced. The best result was when only two classes were chosen for classification, reaching a validation accuracy of 85%.

A customised Vision Transformer for accurate detection and classification of Java Plum leaf disease

Vision Transformer (ViT) has recently attracted significant attention for its performance in image classification. However, few studies have explored its potential in detecting and classifying plant leaf disease. Most existing research on diseased plant leaf detection has focused on non-transformer convolutional neural networks (CNN). Moreover, the studies that applied ViT narrowly experimented using hyperparameters such as image size, patch size, learning rate, attention head, epoch, and batch size. However, these hyperparameters significantly contribute to the model performance. Recognising the gap, this study applied ViT to Java Plum disease detection using optimised hyperparameters. To harness the performance of ViT, this study presents an experiment on Java Plum leaf disease detection. Java Plum leaf diseases significantly threaten agricultural productivity by negatively impacting yield and quality. Timely detection and diagnosis are essential for successful crop management. The primary dataset collected in Bangladesh includes six classes, ‘Bacterial Spot’, ‘Brown Blight’, ‘Powdery Mildew’, and ‘Sooty Mold’, ‘healthy’, and ‘dry’. This experiment contributes to a thorough understanding of Java Plum leaf diseases. Following rigorous testing and refinement, our model demonstrated a significant accuracy rate of 97.51%. This achievement demonstrates the possibilities of using deep-learning tools in agriculture and inspires further research and application in this field. Our research offers a foundational model to ensure crop quality by precise detection, instilling confidence in the global Java Plum market.

The Effect of Meta-heuristic Methods on the Performance of ImageClassification

The Effect of Meta-heuristic Methods on the Performance of ImageClassification

LitefusionNet: Boosting the performance for medical image classification with an intelligent and lightweight feature fusion network

Medical image analysis plays a crucial role in modern healthcare for accurate diagnosis and treatment. However, the inherent challenges and limitations posed by the complexity and variability of medical images, coupled with the shortcomings of existing methods, necessitate the development of novel approaches. In this study, we propose LiteFusionNet (Lightweight Fusion Network), a lightweight model that effectively addresses these challenges, offering the advantage of accurate and efficient medical image classification while mitigating the computational demands. The LitefusionNet leverages the power of deep convolutional neural networks (DCNNs) and feature fusion techniques to achieve improved performance in medical image classification. LitefusionNet combines the strengths of MobileNet and MobileNetV2 architectures to extract robust features from medical images. These features capture discriminative information from different levels of abstraction, enhancing the model's ability to capture fine-grained patterns. The fusion process employs a concatenation method to combine the extracted features, resulting in a more comprehensive representation that improves the model's classification accuracy. To evaluate the effectiveness of LitefusionNet, extensive experiments are conducted on a diverse set of publicly available medical image datasets, including brain MRI, skin, CT, X-ray, and histology. The results demonstrate that LitefusionNet outperforms several existing models in terms of classification accuracy, showcasing its efficacy in different medical imaging modalities. Furthermore, we provide interpretability to the model's predictions by performing Grad-CAM analysis, enabling insights into the important regions in the medical images that contribute to the classification decision. In addition, we compare LitefusionNet with five pre-trained models. LiteFusionNet excels in medical image classification, boasting impressive accuracies across diverse datasets: 97.33% for brain MRI, 91.11% for skin, 99.00% for CT, 98.15% for X-ray, and 92.11% for histology. These results underscore LiteFusionNet's robust and versatile performance, making it a compelling solution for accurate and efficient medical image analysis. Overall, LitefusionNet strikes a balance between accuracy, efficiency, and real-time performance. Our findings demonstrate its potential as a promising solution for accurate and efficient medical image analysis, with applications in diagnostic support systems and clinical decision-making.

Almond (Prunus dulcis) varieties classification with genetic designed lightweight CNN architecture

AbstractAlmond (Prunus dulcis) is a nutritious food with a rich content. In addition to consuming as food, it is also used for various purposes in sectors such as medicine, cosmetics and bioenergy. With all these usages, almond has become a globally demanded product. Accurately determining almond variety is crucial for quality assessment and market value. Convolutional Neural Network (CNN) has a great performance in image classification. In this study, a public dataset containing images of four different almond varieties was created. Five well-known and light-weight CNN models (DenseNet121, EfficientNetB0, MobileNet, MobileNet V2, NASNetMobile) were used to classify almond images. Additionally, a model called 'Genetic CNN', which has its hyperparameters determined by Genetic Algorithm, was proposed. Among the well-known and light-weight CNN models, NASNetMobile achieved the most successful result with an accuracy rate of 99.20%, precision of 99.21%, recall of 99.20% and f1-score of 99.19%. Genetic CNN outperformed well-known models with an accuracy rate of 99.55%, precision of 99.56%, recall of 99.55% and f1-score of 99.55%. Furthermore, the Genetic CNN model has a relatively small size and low test time in comparison to other models, with a parameter count of only 1.1 million. Genetic CNN is suitable for embedded and mobile systems and can be used in real-life solutions.

All-optical Fourier Convolutional Neural Network with High Performance in Image Classification

All-optical Fourier Convolutional Neural Network with High Performance in Image Classification

The use of intelligent real-time image acquisition system using data mining technology in news acquisition and editing practice

Aiming to address the timely dissemination of news information, this work explores the clever utilization of data mining (DM) technology and deep learning (DL) algorithms to construct an intelligent real-time news image acquisition system to meet the urgency of news dissemination needs. First, this work introduces an intelligent real-time news image acquisition system and provides a detailed analysis of its principles and advantages. Throughout this process, the crucial role of DM technology in news image classification and automation is emphasized, especially in dealing with rapidly evolving news events. Next, the work establishes an intelligent real-time news image acquisition model based on DL algorithms, which integrates the essence of DM technology. Through this fusion, the research objective is to enhance the performance of the news image acquisition system to achieve higher real-time and accuracy, which is vital for the swift delivery of news information. Finally, this work investigates the application of the intelligent news image acquisition system in network communication to ensure its adaptability to various network communication scenarios while maintaining accuracy and real-time capabilities. The research results demonstrate that the application of DM technology in combination with DL algorithms can effectively meet the practical needs of the news industry, enhancing the automation of news image processing and enabling faster information delivery to the audience. Notably, the AlexNet model employed performs exceptionally well, achieving recognition rates of up to 99.6% after data augmentation or equalization processing, with an accuracy of 90.9% and a high specificity of 93.38%. This indicates outstanding overall classification accuracy and negative class accuracy, even when distinguishing between news and non-news scenarios. These results clearly underscore the connection between DM technology and news acquisition and editing practices, and emphasize its potential to improve the efficiency and accuracy of real-time information dissemination. The research’s contribution and innovation lie in the fusion of DM technology with DL algorithms to build an intelligent real-time news image acquisition system. This fusion enhances the automation and classification performance of news images, thereby improving the real-time and accuracy of news information. Furthermore, the work strongly emphasizes improving the real-time and accuracy of the news image acquisition system to ensure the swift delivery of information, which is of utmost importance in rapidly changing news events.

Boundary sample-based class-weighted semi-supervised learning for malignant tumor classification of medical imaging.

Medical image classification plays a pivotal role within the field of medicine. Existing models predominantly rely on supervised learning methods, which necessitate large volumes of labeled data for effective training. However, acquiring and annotating medical image data is both an expensive and time-consuming endeavor. In contrast, semi-supervised learning methods offer a promising approach by harnessing limited labeled data alongside abundant unlabeled data to enhance the performance of medical image classification. Nonetheless, current methods often encounter confirmation bias due to noise inherent in self-generated pseudo-labels and the presence of boundary samples from different classes. To overcome these challenges, this study introduces a novel framework known as boundary sample-based class-weighted semi-supervised learning (BSCSSL) for medical image classification. Our method aims to alleviate the impact of intra- and inter-class boundary samples derived from unlabeled data. Specifically, we address reliable confidential data and inter-class boundary samples separately through the utilization of an inter-class boundary sample mining module. Additionally, we implement an intra-class boundary sample weighting mechanism to extract class-aware features specific to intra-class boundary samples. Rather than discarding such intra-class boundary samples outright, our approach acknowledges their intrinsic value despite the difficulty associated with accurate classification, as they contribute significantly to model prediction. Experimental results on widely recognized medical image datasets demonstrate the superiority of our proposed BSCSSL method over existing semi-supervised learning approaches. By enhancing the accuracy and robustness of medical image classification, our BSCSSL approach yields considerable implications for advancing medical diagnosis and future research endeavors.

CONVCAT: A NEW CLASSIFICATION APPROACH USING UC MERCED AND RESISC45 DATASETS

With advances in Earth observation systems, the importance of remote sensing data is increasing daily. These data are used in various fields ranging from image segmentation to terrain classification, from disaster impact assessment to climate change analysis. The use of remotely sensed images for terrain classification has been the subject of a number of studies. This study proposes a new method for terrain classification in the UC Merced Land Use Dataset and RESISC45 remote sensing images. This method is called ConvCat model, which is a combination of classical convolutional layer and CatBoost models. The performance of this model is measured in terms of accuracy, the Matthews Correlation Coefficient (MCC) and the Cohen's Kappa metrics. The results are compared with ensemble models (XGBoost, CatBoost), with ConvXGB, a combination of convolutional learning and XGBoost, and with ResNet50, one of the most widely used transfer learning models. The developed ConvCat model outperformed the other models, achieving an accuracy of 97.44% on the UC Merced data set and an accuracy of 96.89% on the Resisc45 data set. This study shows that our newly developed model provides the best results for the classification problem based on remote sensing images.

IMAGE CLASSIFICATION OF VINE LEAF DISEASES USING COMPLEX-VALUED NEURAL NETWORK

Leaf diseases are a serious challenge in the agricultural industry affecting crop quality and yield especially in grapevines. Early recognition and classification of grape leaf diseases is crucial to enable farmers to take appropriate preventive measures in maintaining the health of their crops. The research utilized an innovative approach based on Complex-Valued Neural Network (CVNN) to address the problem. Using Complex-Valued Neural Network (CVNN) this research seeks to identify and classify grape leaf diseases through a series of experiments. A total of 100 images divided into 4 classes namely Black Rot, ESCA, Leaf Blight, and Healthy were collected to train the model. The results show that the trained CVNN model successfully achieved a training accuracy of 100% and a testing accuracy of 97%, demonstrating excellent performance in classifying grape leaf diseases. This states that the proposed approach has great potential to be an effective tool in helping growers manage their vineyards more efficiently and effectively. The developed image processing method is expected to be applied in designing a system to perform image classification of diseases on grape leaves.

Hybrid quantum neural network structures for image multi-classification

Image classification is a fundamental problem in computer vision, and neural networks provide an effective solution. With the advancement of quantum technology, quantum neural networks have attracted a lot of attention. However, they are only suitable for low-dimensional data and require dimensionality reduction and quantum encoding. Two image classification methods have recently emerged: one employs PCA for dimensionality reduction and angular encoding, and the other integrates QNN into a CNN to improve performance. Despite numerous algorithms, the differences between them remain unclear. This study explores these algorithms’ performance in multi-class image classification and proposes an optimized hybrid quantum neural network suitable for the current environment. As the number of classes increases, research on PCA-based quantum algorithms reveals the barren plateau problem of QNN, which is not suitable for multiple classes in a hybrid setting. Our proposed model combining traditional CNN with QNN addresses QNN’s multi-class training difficulties to some extent and achieves satisfactory classification results. Nevertheless, its accuracy remains below that of the top-performing CNN models. Furthermore, we investigate transfer learning in hybrid quantum neural network models and assess the performance of our models on the quantum hardware from IBM. In conclusion, quantum neural networks show promise but require further research and optimization, facing challenges ahead.

Domain Generalization with Vital Phase Augmentation

Deep neural networks have shown remarkable performance in image classification. However, their performance significantly deteriorates with corrupted input data. Domain generalization methods have been proposed to train robust models against out-of-distribution data. Data augmentation in the frequency domain is one of such approaches that enable a model to learn phase features to establish domain-invariant representations. This approach changes the amplitudes of the input data while preserving the phases. However, using fixed phases leads to susceptibility to phase fluctuations because amplitudes and phase fluctuations commonly occur in out-of-distribution. In this study, to address this problem, we introduce an approach using finite variation of the phases of input data rather than maintaining fixed phases. Based on the assumption that the degree of domain-invariant features varies for each phase, we propose a method to distinguish phases based on this degree. In addition, we propose a method called vital phase augmentation (VIPAug) that applies the variation to the phases differently according to the degree of domain-invariant features of given phases. The model depends more on the vital phases that contain more domain-invariant features for attaining robustness to amplitude and phase fluctuations. We present experimental evaluations of our proposed approach, which exhibited improved performance for both clean and corrupted data. VIPAug achieved SOTA performance on the benchmark CIFAR-10 and CIFAR-100 datasets, as well as near-SOTA performance on the ImageNet-100 and ImageNet datasets. Our code is available at https://github.com/excitedkid/vipaug.

Learning rate burst for superior SGDM and AdamW integration

Current mainstream deep learning optimization algorithms can be classified into two categories: non-adaptive optimization algorithms, such as Stochastic Gradient Descent with Momentum (SGDM), and adaptive optimization algorithms, like Adaptive Moment Estimation with Weight Decay (AdamW). Adaptive optimization algorithms for many deep neural network models typically enable faster initial training, whereas non-adaptive optimization algorithms often yield better final convergence. Our proposed Adaptive Learning Rate Burst (Adaburst) algorithm seeks to combine the strengths of both categories. The update mechanism of Adaburst incorporates elements from AdamW and SGDM, ensuring a seamless transition between the two. Adaburst modifies the learning rate of the SGDM algorithm based on a cosine learning rate schedule, particularly when the algorithm encounters an update bottleneck, which is called learning rate burst. This approach helps the model to escape current local optima more effectively. The results of the Adaburst experiment underscore its enhanced performance in image classification and generation tasks when compared with alternative approaches, characterized by expedited convergence and elevated accuracy. Notably, on the MNIST, CIFAR-10, and CIFAR-100 datasets, Adaburst attained accuracies that matched or exceeded those achieved by SGDM. Furthermore, in training diffusion models on the DeepFashion dataset, Adaburst achieved convergence in fewer epochs than a meticulously calibrated AdamW optimizer while avoiding abrupt blurring or other training instabilities. Adaburst augmented the final training set accuracy on the MNIST, CIFAR-10, and CIFAR-100 datasets by 0.02%, 0.41%, and 4.18%, respectively. In addition, the generative model trained on the DeepFashion dataset demonstrated a 4.62-point improvement in the Frechet Inception Distance (FID) score, a metric for assessing generative model quality. Consequently, this evidence suggests that Adaburst introduces an innovative optimization algorithm that simultaneously updates AdamW and SGDM and incorporates a learning rate burst mechanism. This mechanism significantly enhances deep neural networks’ training speed and convergence accuracy.

Evasion Attacks on Deep Learning-Based Helicopter Recognition Systems

Identifying objects in surveillance and reconnaissance systems with the human eye can be challenging, underscoring the growing importance of employing deep learning models for the recognition of enemy weapon systems. These systems, leveraging deep neural networks known for their strong performance in image recognition and classification, are currently under extensive research. However, it is crucial to acknowledge that surveillance and reconnaissance systems utilizing deep neural networks are susceptible to vulnerabilities posed by adversarial examples. While prior adversarial example research has mainly utilized publicly available internet data, there has been a significant absence of studies concerning adversarial attacks on data and models specific to real military scenarios. In this paper, we introduce an adversarial example designed for a binary classifier tasked with recognizing helicopters. Our approach generates an adversarial example that is misclassified by the model, despite appearing unproblematic to the human eye. To conduct our experiments, we gathered real attack and transport helicopters and employed TensorFlow as the machine learning library of choice. Our experimental findings demonstrate that the average attack success rate of the proposed method is 81.9%. Additionally, when epsilon is 0.4, the attack success rate is 90.1%. Before epsilon reaches 0.4, the attack success rate increases rapidly, and then we can see that epsilon increases little by little thereafter.

Pattern classification of interstitial lung diseases from computed tomography images using a ResNet-based network with a split-transform-merge strategy and split attention.

In patients with interstitial lung disease (ILD), accurate pattern assessment from their computed tomography (CT) images could help track lung abnormalities and evaluate treatment efficacy. Based on excellent image classification performance, convolutional neural networks (CNNs) have been massively investigated for classifying and labeling pathological patterns in the CT images of ILD patients. However, previous studies rarely considered the three-dimensional (3D) structure of the pathological patterns of ILD and used two-dimensional network input. In addition, ResNet-based networks such as SE-ResNet and ResNeXt with high classification performance have not been used for pattern classification of ILD. This study proposed a SE-ResNeXt-SA-18 for classifying pathological patterns of ILD. The SE-ResNeXt-SA-18 integrated the multipath design of the ResNeXt and the feature weighting of the squeeze-and-excitation network with split attention. The classification performance of the SE-ResNeXt-SA-18 was compared with the ResNet-18 and SE-ResNeXt-18. The influence of the input patch size on classification performance was also evaluated. Results show that the classification accuracy was increased with the increase of the patch size. With a 32 × 32 × 16 input, the SE-ResNeXt-SA-18 presented the highest performance with average accuracy, sensitivity, and specificity of 0.991, 0.979, and 0.994. High-weight regions in the class activation maps of the SE-ResNeXt-SA-18 also matched the specific pattern features. In comparison, the performance of the SE-ResNeXt-SA-18 is superior to the previously reported CNNs in classifying the ILD patterns. We concluded that the SE-ResNeXt-SA-18 could help track or monitor the progress of ILD through accuracy pattern classification.

A High-Precision Position Estimation Method for Low Saliency Ratio Machine Based on Convolutional Neural Networks

Sensorless control of surface permanent magnet synchronous motor (SPMSM) faces great challenges in achieving high precision of rotor position estimation at low speed because the saliency of an SPMSM is only a minor secondary effect. In order to extract accurate position information for SPMSM with low and ultra-low saliency ratios, this paper presents a method that implements image recognition techniques with high-frequency (HF) rotating voltage vector injection. For building an image of the induced current vector, the information of the negative-sequence and the second harmonic of positive-sequence HF current signals in the stationary reference frame is extracted. Then, convolutional neural networks (CNN) are applied to the current vector images to establish the relationship between the vector image and the rotor position. Subsequently, the rotor position can be obtained from the result of image recognition from the neural network. Benefiting from CNN’s excellent performance in image classification, the proposed method is able to recognize subtle position signatures even under low saliency. This confirms the opportunity to increase position estimation accuracy for sensorless control of SPMSM. Finally, the experimental results have provided proof that the proposed method has significantly enhanced estimation accuracy.

Bag-Based Feature-Class Correlation Analysis for Multi-Instance Learning Application

Multi-instance Learning (MIL) is widely applied in image classification. In MIL, an image is presented as a bag. A bag consists of multi-instance which is known as patches. Irrelevant features of the image presented to the classifier affects the classification performance. Feature selection is one of the essential phases to select relevant. However, limited studies discuss the feature selection phase in MIL. Correlation between feature-class (FC) relationship is one important criterion to analyse features’ relevance. However, it cannot be performed directly in MIL. To address this gap, this study proposed the MultiBag-FCCorr feature selection technique. It consists of three steps: transformation, evaluation and fusion. The bags of feature information are acquired from summarization from different statistical central tendency measures of trimmed mean, mode and median. In feature evaluation step, extended point biserial correlation has been used to measure FC correlation and then the FC score has been analysed. The selected features are validated via two prominent classifiers (Support Vector Machine (SVM) and K-Nearest Neighbour (KNN)) on benchmark MI image datasets: UCSB Breast Cancer, Tiger, Elephant and Fox datasets. The selected features of UCSB Breast Cancer dataset were reduced to 92% number of features from the proposed technique giving the best result of average accuracy with 86.8.% using SVM and 84.5% using KNN. The average accuracy improved 6.3% using SVM and 16.4% using KNN compared without implementing the proposed feature selection. The results proved that the selected feature set improved the performance of MI image classification.

Hyper-S3NN: Spatial–spectral spiking neural network for hyperspectral image classification

Recently, convolutional neural networks (CNNs) have achieved an appealing performance in hyperspectral image (HSI) classification. Although CNNs excel at capturing local details, they fall short in modeling long-range dependencies due to small receptive fields. Spiking neural networks (SNNs) simulate the encoding and processing of information by the human brain, increasing the receptive field when extracting image features, which is expected to solve the previous problems. Therefore, we present further study on exploiting the power of SNNs for axial global information interaction. The proposed approach, termed Hyper-S3NN, combines CNNs and SNNs to leverage the fine localization information and global contexts. Specifically, based on the leaky integrate-and-fire (LIF) neurons in SNNs, three types of LIF branches are proposed to obtain the global interactive features along the horizontal, vertical, and channel dimensions. By combining convolutions for local information extraction and feature fusion, our Spatial–Spectral LIF (SSLIF) module can effectively integrate these three-dimensional features and enhance the local detail representation capability of the model. Extensive experiments conducted on four benchmark datasets for HSI task (Pavia University, Indian Pines, Trento and Houston 2013) indicate that our method achieves state-of-the-art (SOTA) performance compared with previous methods, which demonstrates the superiority and robust generalization capability of our Hyper-S3NN.

Benchmarking the influence of pre-training on explanation performance in MR image classification.

Convolutional Neural Networks (CNNs) are frequently and successfully used in medical prediction tasks. They are often used in combination with transfer learning, leading to improved performance when training data for the task are scarce. The resulting models are highly complex and typically do not provide any insight into their predictive mechanisms, motivating the field of "explainable" artificial intelligence (XAI). However, previous studies have rarely quantitatively evaluated the "explanation performance" of XAI methods against ground-truth data, and transfer learning and its influence on objective measures of explanation performance has not been investigated. Here, we propose a benchmark dataset that allows for quantifying explanation performance in a realistic magnetic resonance imaging (MRI) classification task. We employ this benchmark to understand the influence of transfer learning on the quality of explanations. Experimental results show that popular XAI methods applied to the same underlying model differ vastly in performance, even when considering only correctly classified examples. We further observe that explanation performance strongly depends on the task used for pre-training and the number of CNN layers pre-trained. These results hold after correcting for a substantial correlation between explanation and classification performance.