Maximum Pooling Research Articles

Chaotic Time Series Recognition: A Deep Learning Model Inspired by Complex Systems Characteristics

A deep learning method is developed for chaotic time series classification. We investigate the chaotic state of a dynamical system, based on the output of the system. One of the main obstacles in time series classification is mapping a high-dimensional vector into a scalar value. To reduce the dimensions, it is common to use an average pooling layer block after feature extraction block. This blind process results in models with high computational complexity and potent to overfitting. One alternative is to extract the features manually, then apply shallow learning models to classify the time series. In fact, since complexity lies between the chaos and order, it is a sound idea to refer to complex systems characteristics to explore the chaotic region entrance. Therefore, chaotic state of a dynamical system can be recognized solely based on these characteristics. Inspired by this concept, we conclude that there is a feature space in which the output vector can be sparsified. Thus, we propose a deep learning method which the feature space dimensions successively are reduced in the feature extraction process. Specifically, we employ a fully convolutional network and add on two maximum pooling layers to the relevant feature extraction block. To validate the proposed model, the Lorenz system is employed which exhibits chaotic/non-chaotic states. We generate a labeled dataset containing 10000 samples each with 20000 features of the output of Lorenz system. The proposed model achieves 99.45 percent accuracy over 2000 unseen samples, higher than all the other competitor methods.

CAT: Learning to collaborate channel and spatial attention from multi‐information fusion

AbstractChannel and spatial attention mechanisms have proven to provide an evident performance boost of deep convolution neural networks. Most existing methods focus on one or run them parallel (series), neglecting the collaboration between the two attentions. In order to better establish the feature interaction between the two types of attentions, a plug‐and‐play attention module is proposed, which is termed as ‘CAT’—activating the Collaboration between spatial and channel Attentions based on learned Traits. Specifically, traits are represented as trainable coefficients (i.e. colla‐factors) to adaptively combine contributions of different attention modules to fit different image hierarchies and tasks better. Moreover, the global entropy pooling is proposed apart from global average pooling and global maximum pooling (GMP) operators, which is an effective component in suppressing noise signals by measuring the information disorder of feature maps. A three‐way pooling operation is introduced into attention modules and the adaptive mechanism is applied to fuse their outcomes. Extensive experiments on MS COCO, Pascal‐VOC, Cifar‐100, and ImageNet show that our CAT outperforms the existing state‐of‐the‐art attention mechanisms in object detection, instance segmentation, and image classification. The model and code will be released soon.

Lightning Identification Method Based on Deep Learning

In this study, a deep learning method called Lightning-SN was developed and used for cloud-to-ground (CG) lightning identification. Based on artificial scenarios, this network model selects radar products that exhibit characteristic factors closely related to lightning. Advanced time of arrival and direction lightning positioning data were used as the labeling factors. The Lightning-SN model was constructed based on an encoder–decoder structure with 25 convolutional layers, five pooling layers, five upsampling layers, and a sigmoid activation function layer. Additionally, the maximum pooling index method was adopted in Lightning-SN to avoid characteristic boundary information loss in the pooling process. The gradient harmonizing mechanism was used as the loss function to improve the model performance. The evaluation results showed that the Lightning-SN improved the segmentation accuracy of the CG lightning location compared with the traditional threshold method, according to the 6-minute operating period of the current S-band Doppler radar, exhibiting a better performance in terms of lightning location identification based on high-resolution radar data. The model was applied to the Ningbo area of Zhejiang Province, China. It was applied to the lightning hazard prevention in the hazardous chemical park in Ningbo. The composite reflectivity and radial velocity were the two dominant factors, with a greater influence on the model performance than other factors.

Image Super-Resolution Network Based on Feature Fusion Attention

The residual structure may learn the entire input region indiscriminately because the residual connection can still learn well as the network depth grows. To a certain extent, the attention mechanism can focus the network’s attention to the interesting area, enhancing the learning performance of essential areas while decreasing the computational load for the system. As a result, the combination of these two advantages could have substantial research significance, for both improve the efficiency and reduce the computational load. A dense residual connection network that combine feature fusion attention approach in image super resolution process is proposed. The dense residual block is enhanced with pixel and channel attention blocks, and a dual-channel path design incorporating global maximum pooling and global average pooling is utilized. A hybrid loss function is also proposed in order to increase the network’s sensitivity to the maximum error between individual pixels. The PSNR/SSIM/ L ∞ performance metrics increased after applying the hybrid loss function and our attention techniques. The experimental results demonstrated that our novel approach has several advantages over some recent approaches, as well as showing good outcomes on many testing datasets.

Sugarcane stem node detection and localization for cutting using deep learning.

In order to promote sugarcane pre-cut seed good seed and good method planting technology, we combine the development of sugarcane pre-cut seed intelligent 0p99oposeed cutting machine to realize the accurate and fast identification and cutting of sugarcane stem nodes. In this paper, we proposed an algorithm to improve YOLOv4-Tiny for sugarcane stem node recognition. Based on the original YOLOv4-Tiny network, the three maximum pooling layers of the original YOLOv4-tiny network were replaced with SPP (Spatial Pyramid Pooling) modules, which fuse the local and global features of the images and enhance the accurate localization ability of the network. And a 1×1 convolution module was added to each feature layer to reduce the parameters of the network and improve the prediction speed of the network. On the sugarcane dataset, compared with the Faster-RCNN algorithm and YOLOv4 algorithm, the improved algorithm yielded an mean accuracy precision (MAP) of 99.11%, a detection accuracy of 97.07%, and a transmission frame per second (fps) of 30, which can quickly and accurately detect and identify sugarcane stem nodes. In this paper, the improved algorithm is deployed in the sugarcane stem node fast identification and dynamic cutting system to achieve accurate and fast sugarcane stem node identification and cutting in real time. It improves the seed cutting quality and cutting efficiency and reduces the labor intensity.

Playing Behavior Classification of Group-Housed Pigs Using a Deep CNN-LSTM Network

The swine industry is one of the industries that progressively incorporates smart livestock farming (SLF) to monitor the grouped-housed pigs’ welfare. In recent years, pigs’ positive welfare has gained much attention. One of the evident behavioral indicators of positive welfare is playing behaviors. However, playing behavior is spontaneous and temporary, which makes the detection of playing behaviors difficult. The most direct method to monitor the pigs’ behaviors is a video surveillance system, for which no comprehensive classification framework exists. In this work, we develop a comprehensive pig playing behavior classification framework and build a new video-based classification model of pig playing behaviors using deep learning. We base our deep learning framework on an end-to-end trainable CNN-LSTM network, with ResNet34 as the CNN backbone model. With its high classification accuracy of over 92% and superior performances over the existing models, our proposed model highlights the importance of applying the global maximum pooling method on the CNN final layer’s feature map and leveraging a temporal attention layer as an input to the fully connected layer for final prediction. Our work has direct implications on advancing the welfare assessment of group-housed pigs and the current practice of SLF.

Machine Vision Product Appearance Defect Detection Based on Deep Learning

Due to the influence of objective environmental conditions in the image acquisition stage, the image quality of the existing methods is poor, which makes the effect of defect detection not ideal. In this paper, research on product appearance defect detection based on deep learning machine vision is proposed. A product appearance model is constructed by using a Gaussian mixture, and multiple sub-images with different sizes are used as the data support of the multi-scale expression of the Gaussian model to calculate the Gaussian difference features of the image. A Faster R-CNN network algorithm in deep learning is used to recognize defects. The multi-task mechanism is introduced in the setting of the loss function to realize the concurrent function of defect detection. The mean pooling and maximum pooling operations are introduced in the setting of the classification regression network to realize the comprehensive detection of target images. Judge whether that product has a defect according to the output result of the algorithm and the DOG characteristics of the image. The test results show that the accuracy of the designed method for various types of defect detection is more than 90. 00% and it has a high detection efficiency.

A Person Re-Identification Method with Multi-Scale and Multi-Feature Fusion

In order to solve the problem of insufficient expression of pedestrian features and ignoring the semantic relevance of adjacent areas in person re-identification, a multi-scale and multi-feature fusion person re-identification model is proposed. Firstly, feature maps containing different information are obtained by multi-branch structure. Secondly, the semantic correlation between local features is emphasized by combining adjacent local regions. Finally, combining the advantages of maximum pooling and average pooling, the model can learn more comprehensive feature information from different directions. Experiments are carried out on Market-1501, DukeMTMC-reID and MSMT17 respectively, it can be evaluated that the model can effectively extract pedestrian feature and has high recognition accuracy under different lighting and camera angles.

Multilevel hybrid accurate handcrafted model for myocardial infarction classification using ECG signals.

Myocardial infarction (MI) is detected using electrocardiography (ECG) signals. Machine learning (ML) models have been used for automated MI detection on ECG signals. Deep learning models generally yield high classification performance but are computationally intensive. We have developed a novel multilevel hybrid feature extraction-based classification model with low time complexity for MI classification. The study dataset comprising 12-lead ECGs belonging to one healthy and 10 MI classes were downloaded from a public ECG signal databank. The model architecture comprised multilevel hybrid feature extraction, iterative feature selection, classification, and iterative majority voting (IMV). In the hybrid handcrafted feature (HHF) generation phase, both textural and statistical feature extraction functions were used to extract features from ECG beats but only at a low level. A new pooling-based multilevel decomposition model was presented to enable them to create features at a high level. This model used average and maximum pooling to create decomposed signals. Using these pooling functions, an unbalanced tree was obtained. Therefore, this model was named multilevel unbalanced pooling tree transformation (MUPTT). On the feature extraction side, two extractors (functions) were used to generate both statistical and textural features. To generate statistical features, 20 commonly used moments were used. A new, improved symmetric binary pattern function was proposed to generate textural features. Both feature extractors were applied to the original MI signal and the decomposed signals generated by the MUPTT. The most valuable features from among the extracted feature vectors were selected using iterative neighborhood component analysis (INCA). In the classification phase, a one-dimensional nearest neighbor classifier with ten-fold cross-validation was used to obtain lead-wise results. The computed lead-wise results derived from all 12 leads of the same beat were input to the IMV algorithm to generate ten voted results. The most representative was chosen using a greedy technique to calculate the overall classification performance of the model. The HHF-MUPTT-based ECG beat classification model attained excellent performance, with the best lead-wise accuracy of 99.85% observed in Lead III and 99.94% classification accuracy using the IMV algorithm. The results confirmed the high MI classification ability of the presented computationally lightweight HHF-MUPTT-based model.

Lightweight Multi-Attention Fusion Network for Image Super-Resolution

Single image super-resolution reconstruction (SISR) is one of the important techniques in computer vision and image processing. Most of the existing SISR methods adopt equal processing for different spatial domains and channel domains, resulting in a large amount of computational resources wasted on unimportant features. In order to address these problems, a novel lightweight multi-attention fusion network (LMAFN) is proposed, in which the multiple attention fusion block allocates computational resources more efficiently by capturing the weight information implied by the channel domain and the spatial domain separately, thus effectively reducing the number of parameters. The synthetic channel attention block in the multiple attention fusion block makes full use of inter-channel correlation by introducing both global standard deviation pooling and maximum pooling. Global features are fused through residual linking to alleviate the problem of high frequency information loss. Experimental results on several benchmark datasets show that the proposed method effectively reduces the number of parameters and computational effort without excessive loss of reconstruction performance, and achieves better performance than the compared models.

A lightweight multiscale target object detection network for melanoma based on attention mechanism manipulation

To propose a deep learning target detection model AM- YOLO that integrates coordinate attention and efficient attention mechanism. Mosaic image enhancement and MixUp mixed-class enhancement were used for image preprocessing. In the target detection model YOLOv5s with One-Stage structure and modified backbone network and neck network, the maximum pooling layer of the spatial pyramid of the backbone network was replaced with a two-dimensional maximum pooling layer, and the coordinate attention mechanism and the efficient channel attention mechanism were integrated into the C3 module and the backbone network of the model, respectively. The improved model was compared with the unmodified YOLOv5s model, YOLOv3 model, YOLOv3-SPP model, and YOLOv3-tiny model for relevant algorithmic indicators in comparative experiments. The AM-YOLO model incorporating coordinate attention and efficient channel attention mechanism effectively improved the accuracy of melanoma recognition with also a reduced size of the model weight. This model showed significantly better performance than other models in terms of precision, recall rate and mean average precision, and its mean average precision for benign and malignant melanoma reached 92.8% and 87.1%, respectively. The deep learning-based target object detection algorithm model can be applied in recognition of melanoma targets.

Combining Max pooling-Laplacian theory and k-means clustering for novel camouflage pattern design.

Camouflage is the main means of anti-optical reconnaissance, and camouflage pattern design is an extremely important step in camouflage. Many scholars have proposed many methods for generating camouflage patterns. k-means algorithm can solve the problem of generating camouflage patterns quickly and accurately, but k-means algorithm is prone to inaccurate convergence results when dealing with large data images leading to poor camouflage effects of the generated camouflage patterns. In this paper, we improve the k-means clustering algorithm based on the maximum pooling theory and Laplace's algorithm, and design a new camouflage pattern generation method independently. First, applying the maximum pooling theory combined with discrete Laplace differential operator, the maximum pooling-Laplace algorithm is proposed to compress and enhance the target background to improve the accuracy and speed of camouflage pattern generation; combined with the k-means clustering principle, the background pixel primitives are processed to iteratively calculate the sample data to obtain the camouflage pattern mixed with the background. Using color similarity and shape similarity for evaluation, the results show that the combination of maximum pooling theory with Laplace algorithm and k-means algorithm can effectively solve the problem of inaccurate results of k-means algorithm in processing large data images. The new camouflage pattern generation method realizes the design of camouflage patterns for different backgrounds and achieves good results. In order to verify the practical application value of the design method, this paper produced test pieces based on the designed camouflage pattern generation method and tested the camouflage effect of camouflage pattern in sunny and cloudy days respectively, and the final test results were good.

Facial Recognition of Cattle Based on SK-ResNet

With the gradual increase of the scale of the breeding industry in recent years, the intelligence level of livestock breeding is also improving. Intelligent breeding is of great significance to the identification of livestock individuals. In this paper, the cattle face images are obtained from different angles to generate the cow face dataset, and a cow face recognition model based on SK_ResNet is proposed. Based on ResNet-50 and using a different number of sk_Bottleneck, this model integrates multiple receptive fields of information to extract facial features at multiple scales. The shortcut connection part connects to the maximum pooling layer to reduce information loss; the ELU activation function is used to reduce the vanishing gradient, prevent overfitting, accelerate the convergence speed, and improve the generalization ability of the model. The constructed bovine face dataset was used to train the SK-ResNet-based bovine face recognition model, and the accuracy rate was 98.42%. The method was tested on the public dataset and the self-built dataset. The accuracy rate of the model was 98.57% on the self-built pig face dataset and the public sheep face dataset. The accuracy rate was 97.02%. The experimental results verify the superiority of this method in practical application, which is helpful for the practical application of animal facial recognition technology in livestock.

A Novel CNN pooling layer for breast cancer segmentation and classification from thermograms.

Breast cancer is the second most frequent cancer worldwide, following lung cancer and the fifth leading cause of cancer death and a major cause of cancer death among women. In recent years, convolutional neural networks (CNNs) have been successfully applied for the diagnosis of breast cancer using different imaging modalities. Pooling is a main data processing step in CNN that decreases the feature maps' dimensionality without losing major patterns. However, the effect of pooling layer was not studied efficiently in literature. In this paper, we propose a novel design for the pooling layer called vector pooling block (VPB) for the CCN algorithm. The proposed VPB consists of two data pathways, which focus on extracting features along horizontal and vertical orientations. The VPB makes the CNNs able to collect both global and local features by including long and narrow pooling kernels, which is different from the traditional pooling layer, that gathers features from a fixed square kernel. Based on the novel VPB, we proposed a new pooling module called AVG-MAX VPB. It can collect informative features by using two types of pooling techniques, maximum and average pooling. The VPB and the AVG-MAX VPB are plugged into the backbone CNNs networks, such as U-Net, AlexNet, ResNet18 and GoogleNet, to show the advantages in segmentation and classification tasks associated with breast cancer diagnosis from thermograms. The proposed pooling layer was evaluated using a benchmark thermogram database (DMR-IR) and its results compared with U-Net results which was used as base results. The U-Net results were as follows: global accuracy = 96.6%, mean accuracy = 96.5%, mean IoU = 92.07%, and mean BF score = 78.34%. The VBP-based results were as follows: global accuracy = 98.3%, mean accuracy = 97.9%, mean IoU = 95.87%, and mean BF score = 88.68% while the AVG-MAX VPB-based results were as follows: global accuracy = 99.2%, mean accuracy = 98.97%, mean IoU = 98.03%, and mean BF score = 94.29%. Other network architectures also demonstrate superior improvement considering the use of VPB and AVG-MAX VPB.

Vehicle Interior Sound Classification Based on Local Quintet Magnitude Pattern and Iterative Neighborhood Component Analysis

ABSTRACT Nowadays, environmental sound classification (ESC) has become one of the most studied research areas. Sound signals that are indistinguishable from the human auditory systems have been classified by computer-aided systems and machine learning methods. Therefore, ESC has been used in signal processing and sound forensics applications. A novel ESC type is presented in this paper, and it is named as vehicle interior sound classification (VISC). VISC is defined as one of the sub-branches of the ESC, and it is utilized as sound-based biometrics for vehicles. A hand-crafted feature-based VISC method is presented. The proposed method has multileveled feature generation by using maximum pooling and the proposed local quintet magnitude pattern (LQMP), feature selection with iterative neighborhood component analysis (INCA), and classification phases. A novel VISC dataset was collected from YouTube and the proposed LQMP and INCA based method applied to the collected sounds. The results denoted that following: the accuracy, F1-score, and geometric mean of the proposed LQMP and INCA based VISC method were calculated as 98.38%,98.23%, and 98.21% by using support vector machine classifier respectively. The contribution of the proposed VISC method is to denote that the vehicles can be classified by using sound.

Research on Surface Defect Detection of Camera Module Lens Based on YOLOv5s-Small-Target

For the problem of low resolution of camera module lens surface defect image, small target and blurred defect details leading to low detection accuracy, a camera module lens surface defect detection algorithm YOLOv5s-Defect based on improved YOLOv5s is proposed. Firstly, to solve the problems arising from the anchor frame generated by the network through K-means clustering, the dynamic anchor frame structure DAFS is introduced in the input stage. Secondly, the SPP-D (Spatial Pyramid Pooling-Defect) improved from the SPP module is proposed. The SPP-D module is used to enhance the reuse rate of feature information in order to reduce the loss of feature information due to the maximum pooling of SPP modules. Then, the convolutional attention module is introduced to the network model of YOLOv5s, which is used to enhance the defective region features and suppress the background region features, thus improving the detection accuracy of small targets. Finally, the post-processing method of non-extreme value suppression is improved, and the improved method DIoU-NMS improves the detection accuracy of small targets in complex backgrounds. The experimental results show that the mean average precision mAP@0.5 of the YOLOv5s-Small-Target algorithm is 99.6%, 8.1% higher than that of the original YOLOv5s algorithm, the detection speed FPS is 80 f/s, and the model size is 18.7M. Compared with the traditional camera module lens surface defect detection methods, YOLOv5s-Small-Target can detect the type and location of lens surface defects more accurately and quickly, and has a smaller model volume, which is convenient for deployment in mobile terminals, meeting the demand for real-time and accuracy of camera module lens surface defect detection.

Spectrum Sensing Using CNN With Attention on Switch of Channel States

This work addresses the issue of spectrum sensing with random arrival and departure of primary signals. We first design a convolutional neural network (CNN) with outputs as the posterior probabilities of the arrival and departure of primary signals, leading to a CNN-based detector with the ratio of the posterior probabilities (i.e., the outputs of the CNN) as a test statistic. To further enhance the attention of the network on the switch feature of channel states, we design a switch attention module (SAM) that adaptively weights the received signals. Replacing the convolution plus maximum pooling block in the CNN detector with the SAM block leads to an SAM-CNN detector. Simulations show that the proposed CNN detector significantly outperforms existing detectors, and further improvement of detection probability by 19% is achieved by the SAM-CNN detector.

Multi-scale convolutional network with channel attention mechanism for rolling bearing fault diagnosis

In recent years, deep learning has achieved great success in bearing fault diagnosis due to its robust feature learning capabilities. However, in the actual industry, the diagnostic accuracy would be degraded under varying operation conditions or in noisy environments. To enhance the diagnostic performance in industrial applications, a Multi-scale Convolutional Neural Network with Channel Attention (CA-MCNN) is proposed in this paper. In CA-MCNN, the maximum pooling and average pooling layers are used to extract the multi-scale information of the bearing signals, which increases the dimensions of input. The channel attention mechanism is introduced to increase the convolutional layer feature learning ability by adaptively scoring and assigning weights to the learned features. Moreover, the feature parallel fusion mechanism based on 1-D convolution is applied to capture complementary multi-scale information and reduce network complexity. The performance of CA-MCNN is compared with other fault diagnosis models, and experimental results verify that the CA-MCNN achieves the highest diagnosis accuracy under noisy environments and varying working speeds.

Research on neural network algorithm in artificial intelligence recognition

With the rapid development of computer application technology, computer artificial intelligence recognition technology has also made rapid progress. Artificial intelligence was first represented by LISP language, machine theorem proof, etc., and then the intelligent medical treatment and neural network appeared, and now it realizes the popularization of AI. The purpose of this article is to explore the application of neural network algorithms in artificial intelligence recognition. Based on the classic convolutional neural network with LeNet structure, this paper constructs a reasonable convolutional neural network model by deepening and expanding, and uses it in face recognition to conduct face recognition research. LeNet is the origin of CNN, and LeNet is mainly used to identify and classify handwritten characters. This paper uses three networks on the YaleB face database for face recognition, and compares the pooling method of the maximum pooling and average pooling, the misclassification rate of the ReLU incentive function and the tanh incentive function on the three networks.YaleB is a very complete face database, divided into five subsets, ten people, and 64 images per person, including changes in illumination, expressions, and gestures. The tanh function has what is an odd function and has a soft saturation property, while the relu function has a hard saturation property. Analyze whether it can solve the problems caused by illumination factors and occlusion factors in traditional face recognition. Finally, the recognition effect is compared with the traditional face recognition method, and it is concluded that the effect of the traditional face recognition method is not as high as the misclassification rate of 0.0356 based on the convolutional neural network. The recognition method works well. Maximum pooling is also known as maximum reduced pixel sampling, and is more based on the input element map than average pooling.Through the optimization of the algorithm, the face recognition rate is improved, which helps the application and development of artificial intelligence.

Adaptive features selection and EDNN based brain image recognition on the internet of medical things

A brain tumor is one of the fundamental explanations behind fatality among different sorts of malignant growth; the brain is a touchy, intricate, and focal part of the body. The appropriate and convenient conclusion can deflect the life of an individual somewhat. Subsequently, inside this research study, we proposed a brain tumor classification design to consolidate Adaptive Feature Selection and Entropy-based Deep Neural Network (EDNN) based brain image recognition on the Internet of Medical Things (IOMT). The framework regularly contains four modules pre-processing, including extraction, feature selection or determination, and tumor classification. Primarily, we take out the commotion that is noise from the image. The de-noised image is moved to the skull stripping for the brain region extraction. Afterward, the functional features, such as SFTA, geometric, and LBP, are extracted from pre-processed release images. Select the best features using the Adaptive Grill Herd (AKH) Learning Optimization algorithm. Finally, the proposed EDNN classifies brain images as normal brain or abnormal brain based on the extracted features. The EDNN classifier is modified with entropy-based normalization to mitigate the excess overlap inside the neural network's deeper layer towards the maximum pooling layer. Experimental grades demonstrate that the proposed scheme achieves enhanced results compared to the obtainable strategies.