Explainable active reinforcement deep learning improves lung cancer detection from CT images.

Feature extraction in electronic nose (E-nose) signals is the first signal processing step in machine olfactory systems. Traditional feature extraction methods are often cumbersome and depend greatly on the experience of the researcher. The failure of feature extraction seriously deteriorates the performance of subsequent pattern recognition algorithms. In addition, inherent sensor signal shifts among sensing systems with identical configurations affect E-nose performance. Therefore, due to the unfavorable performance of empirical feature extraction and poor reproducibility, the odor detection ability of E-nose systems has deteriorated. In this paper, a bionic olfactory signal processing (BOSP) network based on a spiking neural network (SNN) is proposed to address these issues. Inspired by the structure of the olfactory bulb in mammals, the system has two components: a virtual olfactory receptor (VOR) layer and a bionic olfactory bulb (BOB) layer. The VORs convert the sensor response signals into spike sequences through two spike coding strategies. The BOBs, consisting of mitral cells and granule cells, extract spatiotemporal features from the spike sequences. The proposed BOSP network is a feature-level model that automatically extracts useful features from the sensor array outputs without applying traditional and tedious feature extraction steps, which can subsequently cascade various classifiers. The results of extensive experiments demonstrate that the BOSP network not only shows better feature extraction capability than several state-of-the-art methods but also exhibits promising shift suppression ability.

This paper studies the challenging problem of object detection using rich image and depth features. An invariant Hough random ferns framework for RGB-D images is proposed here, which primarily consists of a rotation-invariant RGB-D local binary feature, random ferns classifier training, Hough mapping and voting, searches for the maxima, and back projection. In comparison with traditional three-dimensional local feature extraction techniques, this method is effective in reducing the amount of computation required for feature extraction and matching. Moreover, the detection results showed that the proposed method is robust against rotation and scale variations, changes in illumination, and part-occlusions. The authors believe that this method will facilitate the use of perception in fields such as robotics.

Vehicle type and brand information constitute a crucial element in intelligent transportation systems (ITSs). While numerous appearance-based classification methods have studied frontal view images of vehicles, the challenge of multi-pose and multi-angle vehicle distribution has largely been overlooked. This paper proposes an appearance-based classification approach for multi-angle vehicle information recognition, addressing the aforementioned issues. By utilizing faster regional convolution neural networks, this method automatically captures crucial features for vehicle type and brand identification, departing from traditional handcrafted feature extraction techniques. To extract rich and discriminative vehicle information, ZFNet and VGG16 are employed. Vehicle feature maps are then imported into the region proposal network and classification location refinement network, with the former generating candidate regions potentially containing vehicle targets on the feature map. Subsequently, the latter network refines vehicle locations and classifies vehicle types. Additionally, a comprehensive vehicle dataset, Car5_48, is constructed to evaluate the performance of the proposed method, encompassing multi-angle images across five vehicle types and 48 vehicle brands. The experimental results on this public dataset demonstrate the effectiveness of the proposed approach in accurately classifying vehicle types and brands.

Human motion behavior recognition (HMBR) is a critical research area in computer vision, enabling automated understanding of human activities through deep learning models. This field has witnessed rapid advancements due to the growing availability of large-scale motion datasets and computational improvements in deep learning architectures. Applications range from healthcare monitoring and security surveillance to sports analytics and human-computer interaction (HCI). Traditional handcrafted feature extraction techniques have been largely replaced by deep learning models, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory networks (LSTMs), and transformer-based architectures. This paper presents a comprehensive study on real-time HMBR using deep learning, covering data acquisition, preprocessing, model architectures, optimization techniques, and real-time deployment strategies. Additionally, challenges such as occlusion, noise, computational efficiency, and ethical concerns are discussed, along with future research directions in self-supervised learning, multimodal data fusion, and scalable real-world deployment.

Nonlocal feature learning based on a variational graph auto-encoder network for small area change detection using SAR imagery

In the field of the fault diagnosis of industrial processes, there are many problems in process data, such as missing critical fault data, high repeatability of normal state data, and poor representation of faults data, which may reduce the accuracy of fault diagnosis. In this paper, a novel K-medoids based synthetic minority over-sampling technique that combines locality preserving projections (KMS-LPP), is proposed for fault diagnosis. First, the synthetic minority sampling technology (SMOTE) is designed based on K-medoid to generate minority fault samples to address the imbalanced problem of data. Second, to extract the key fault-relevant features and reserve the local structure information at the same time, the manifold learning (ML) approach locality preserving projections (LPP) is performed to reduce the dimensionality of data. Finally, the Adaboost. M2, as an ensemble classifier, is conducted for fault classification. Simulations of the Tennessee Eastman process (TEP) are performed for validation of the performance of the presented KMS-LPP method. The obtained results show that KMS-LPP has enhanced the performance of fault diagnosis due to its higher accuracy compared with traditional oversampling and feature extraction methods, which indicates the effectiveness of KMS-LPP.

Parkinson’s disease (PD), the second most prevalent neurodegenerative disorder globally, afflicting approximately 10 million individuals, necessitates early detection for optimal management. In this paper, we propose deep learning models to discern Parkinson’s disease through the nuanced analysis of handwriting with the overall objective of achieving transparency and trustworthiness through the integration of Explainable and Interpretable AI. Leveraging transfer learning from well-established VGG16 and VGG19 architectures and introducing two bespoke CNN models (PD-Detect1 and PD-Detect2), we meticulously scrutinize diverse datasets (HandPD, NewhandPD, Parkinson Drawing) to ascertain the efficacy of our approach. LIME and SHAP Explainable AI techniques are employed to pinpoint specific regions of the spiral drawings that significantly influence the predictions made by the VGG16 and PD-Detect2 models. Additionally, Convolutional Filter Visualization and Grad-CAM are utilized to illustrate how the convolutional layers of the PD-Detect2 model function. Finally, LIME is applied to the PD-Detect2 model to identify visual markers of handwriting symptoms in the spiral drawing. Remarkable results underscore our reliance on VGG16 and VGG19 for precise identification, achieving an outstanding 100% accuracy in the waves drawing dataset. PD-Detect1 exhibits commendable performance with an accuracy of 94.44% in the meander of NewhandPD dataset, while VGG16 achieves an accuracy of 95%. VGG16 records 95% accuracy while PD-Detect2 achieves 85% accuracy in the spiral drawing dataset. Both results further bolstered to 100% with the application of classic data augmentation techniques. The Positive/Negative superpixels from LIME and SHAP highlight the key regions used by VGG16 and PD-Detect2 for predictions, while PD-Detect2 places more emphasis on disease-related features. Visualizing convolutional filters provided insight into the functionality of each layer within the PD-Detect2 model, while the Class Activation Maps produced by Grad-CAM highlighted the image regions most influential to the model’s decision. Ultimately, LIME’s superpixels identified visual markers of handwriting symptoms associated with Parkinson’s disease. Explainable AI and Interpretable AI enhance the quality of CNN models and support the decision-making process, enabling healthcare professionals to more accurately assess disease probability and monitor treatment responses. This leads to a more effective system for the early diagnosis of Parkinson’s disease through prediction and visual monitoring of handwriting symptoms.

Signa1 modulation recognition (SMR) is an essential and challenging topic in orthogonal frequency-division multiplexing (OFDM) systems, and also it is the fundamental technique for signal detection and recovery. However, traditional feature extraction based SMR methods cannot effectively acquire the characteristics of the OFDM signals. Hence, the modulated OFD-M signal cannot be reliably identified. In this paper, we propose a deep learning (DL) based SMR method for recognizing OFDM signals, which is combined with a convolutional neural network (CNN) trained on in-phase and quadrature (IQ) samples. In the network model, the batch normalization (BN) layer and dropout layer are used to speed up model training and prevent overfitting, respectively. Three convolution layers with different convolution kernels perform well than traditional feature extraction methods in obtaining intrinsic properties of OFDM signals. The same number of multiple modulated signals are mixed and sent to the trained model for identification. Experiments are conducted to show that the method we proposed performs better than the traditional methods, mainly reflected in a higher probability of correct classification (PCC) and better consistency.

Remote sensing technology has been modernized by artificial intelligence, which has made it possible for deep learning algorithms to extract useful information from images. However, overfitting and lack of uncertainty quantification, high-resolution images, information loss in traditional feature extraction, and background information retrieval for detected objects limit the use of deep learning models in various remote sensing applications. This paper proposes a Bayes by backpropagation (BBB)-based system for scene-driven identification and information retrieval in order to overcome the above-mentioned problems. We present the Bayes R-CNN, a two-stage object detection technique to reduce overfitting while also quantifying uncertainty for each object recognized within a given image. To extract features more successfully, we replace the traditional feature extraction model with our novel Multi-Resolution Extraction Network (MRENet) model. We propose the multi-level feature fusion module (MLFFM) in the inner lateral connection and a Bayesian Distributed Lightweight Attention Module (BDLAM) to reduce information loss in the feature pyramid network (FPN). In addition, our system incorporates a Bayesian image super-resolution model which enhances the quality of the image to improve the prediction accuracy of the Bayes R-CNN. Notably, MRENet is used to classify the background of the detected objects to provide detailed interpretation of the object. Our proposed system is comprehensively trained and assessed utilizing the state-of-the-art DIOR and HRSC2016 datasets. The results demonstrate our system’s ability to detect and retrieve information from remote sensing scene images.

Chapter 10 - Analysis of Histopathology Images: From Traditional Machine Learning to Deep Learning

Visual inspection of target parts is a common approach to ensuring train safety. However, some key parts, such as fastening bolts, do not possess sufficient feature information, because they are usually small, polluted, or obscured. These factors affect inspection accuracy and can lead to serious accidents. Therefore, traditional visual inspection relying on feature extraction cannot always meet the requirements of high‐accuracy inspection. Deep learning has considerable advantages in image recognition for autonomous information mining, but it requires a considerable amount of computation. To resolve the issues mentioned above, this study proposes a method that combines traditional visual inspection with deep learning. Traditional feature extraction is used to locate the targets approximately, which makes the deep learning purposeful and efficient. A composite neural network, stacked auto‐encoder convolutional neural network (SAE‐CNN), is provided to further improve the training efficiency. A SAE is added to a CNN so that the network can obtain optimum results faster and more accurately. Taking the inspection of centre plate bolts in a moving freight car as an example, the overall system and specific processes are described. The study results showed satisfactory accuracy. A related analysis and comparative experiment were also conducted.

This study describes a method for classifying electrocorticograms (ECoGs) based on motor imagery (MI) on the brain–computer interface (BCI) system. This method is different from the traditional feature extraction and classification method. In this paper, the proposed method employs the deep learning algorithm for extracting features and the traditional algorithm for classification. Specifically, we mainly use the convolution neural network (CNN) to extract the features from the training data and then classify those features by combing with the gradient boosting (GB) algorithm. The comprehensive study with CNN and GB algorithms will profoundly help us to obtain more feature information from brain activities, enabling us to obtain the classification results from human body actions. The performance of the proposed framework has been evaluated on the dataset I of BCI Competition III. Furthermore, the combination of deep learning and traditional algorithms provides some ideas for future research with the BCI systems.

Educational data produced by various systems, such as e-learning, student admission process, online student exam, and automated significance management solutions, may be successfully analyzed using educational data mining (EDM) methods to provide beneficial for the academic student’s progress. A highly desired use of EDM is the prediction of standardized testing from previous academic data. In this intelligence, it is critical to establish an automated approach for predicting student success. The majority of existing research on academic prediction problem use traditional feature extraction and illustration approaches, in which retrieved characteristics are feed to machine learning classifier. Supervised learning has recently allowed academics to extract high-level characteristics from raw data automatically. The proposed enhanced feature representation approaches allow for better performance on real time heterogenous data. In this paper, we demonstrate the Long Short-Term Memory (LSTM) with Recurrent Neural Network (RNN) called as RNN-LSTM. By examining current research difficulties based on evolving feature classification and prediction, we applied the most sophisticated LSTM paired with an attentiveness process approach in this work. Academicians, institutions, and governmental organizations benefit from this study since it allows them to anticipate achievement early. In extensive experimental analysis proposed work compared with numerous state-of-the-art, the better recurrent neural network functionalities of LSTM matching with the hybrid mechanism give higher performance. Predictive performance of 90.50% has achieved using the proposed algorithm.

Face recognition has many important applications in areas such as public surveillance and security, identity verification in the digital world, and modeling techniques in multimedia data management. Facial expression recognition is also important for targeted marketing, medical analysis, and human–robot interaction. In this paper, we survey a few techniques for facial analysis. We compare the cloud platform AWS Rekognition, convolutional neural networks, transfer learning from pre-trained neural nets, and traditional feature extraction using facial landmarks for this analysis. Although not comprehensive, this survey covers a lot of ground in the state-of-the-art solutions for facial analysis. We show that to get high accuracy, good-quality data and processing power must be provided in large quantities. We present the results of our experiments which have been conducted over six different public as well as proprietary image data sets.

The power system measurement data has high-dimensional features and strong noise, which is difficult to be directly used for intrusion detection. Traditional feature extraction and selection methods take the feature processing as a preprocessing step and perform separately from the model training, which makes the features not well adapted to the model. Therefore, we propose a novel wrapped feature selection framework based on the Las Vegas algorithm, which can improve the detection accuracy by strengthening the coupling between the feature selection and the model training. The Las Vegas algorithm can evaluate a feature subset through a specified model. In this paper, a heuristic thinking is integrated into the Las Vegas algorithm, which greatly improves the search performance of the original method. Finally, the proposed framework is examined on the IEEE 14-bus, 39-bus and 57-bus test systems for two attacks and the experimental results proves the effectiveness and stability of the proposed framework.