Plant diseases pose a significant danger to world agricultural production, which requires scalable and effective solutions for early prediction and management. The current work explores deep technology that uses learning to identify plant illnesses, leveraging Convolutional Neural Networks (CNN) for the detection and diagnosis of plant disorders in multiple crop species: corn, potato, rice, tea, and tomato. The system adopts a two-model strategy, involving the fine-tuning of a pre-trained ResNet50 and the development of a custom CNN architecture. Our dual-model framework combines a fine-tuned ResNet50 for high-accuracy diagnosis and a lightweight custom CNN optimized for edge deployment, achieving up to 99.33

Schizophrenia (SZ), a complicated mental illness, shows up as incorrect beliefs, perceptual distortions, poor reasoning, and neurocognitive problems. Its varied character—varying in severity, development, and therapy response—offers major diagnostic difficulty. While Magnetic Resonance Imaging (MRI) provides comprehensive structural and functional imaging for a better knowledge of the condition, Electroencephalography (EEG) with its great temporal resolution offers vital insights into neuronal malfunction in SZ. Deep learning (DL) techniques as well as advanced machine learning (ML) have been developed to improve SZ detection, therefore facilitating quick identification and improved patient recovery. This study presents SZAtt-Net, a DL framework designed for SZ detection and classification, integrating Convolutional Neural Networks (CNNs), Bidirectional Gated Recurrent Unit (BiGRU), along with Multilayer Perceptron (MLP) architectures. A key contribution of this work is the comprehensive ablation study of channel, self, spatial, and temporal attention mechanisms in DL, conducted to assess their impact on model performance across multimodal data. Notably, due to the absence of dedicated MRI datasets for SZ classification, this study repurposes an MRI segmentation dataset for classification, making it the first such attempt. As a unified model, SZAtt-Net is applied separately to three benchmark datasets—Kaggle EEG, LMSU EEG, and Hippocampus MRI—achieving accuracy rates of 99.37% and 98.92% using channel attention, and 96.33% using spatial attention, respectively. The proposed framework is also benchmarked against various pre-trained models, and a Gradient-weighted Class Activation Mapping (Grad-CAM) analysis is performed to enhance interpretability. This work underscores the clinical relevance of SZAtt-Net and outlines future research directions for improving accessible and accurate diagnostic solutions for SZ.

In recent years, the growing demand for assistive technology in healthcare, fitness tracking and smart environments has positioned Human Activity Recognition (HAR) as vital in the era of data-driven intelligent systems. Identification of motion patterns from various smartphone and wearable sensors is considered to be critical for developing efficient recognition systems. Traditional approaches tend to struggle in capturing complex dependencies in sensor data which leads us to find hybrid techniques for HAR. In this paper, we design a two-stage deep feature extraction and selection technique termed FEST using image encoding derived from time series signal data from sensors. Here, image encoding using Gramian Angular Matrix has been employed to form activity images from sensor readings. These images were then fed to three well-known pretrained visual models, InceptionResNetV2, Xception, EfficientNetV2B1 to capture complex spatial-temporal relations. The deep features were then extracted for optimal feature selection through a two-step selection strategy involving pre-filter Chi-square followed by Honey-Badger meta-heuristic algorithm. The proposed framework has been examined on three public datasets, UCI-HAR, MHEALTH and, KU-HAR which achieved an overall accuracy of 94.00

Parkinson’s Disease (PD), an enduring and degenerative disorder of the nervous system, presents significant challenges for timely and accurate diagnosis due to its complex and variable symptomatology. In this study, we propose DeepPark-Net, a unified CNN-BiGRU-MLP framework designed to independently process Electroencephalography (EEG) and Magnetic Resonance Imaging (MRI) data for multimodal PD detection. Unlike prior studies, we perform evaluations under both subject-dependent (SD) and, for the first time, subject-independent (SI) settings for EEG, demonstrating the model’s robustness to inter-subject variability. Experimental results show that DeepPark-Net achieves 100 https://github.com/Sankhadip-007/DeepPark-Net-A-Multimodal-Deep-Learning-Framework-for-Parkinson-s-Disease-Detection .

Arrhythmia detection is important for early identification of irregular heart activities to prevent serious complications like stroke, cardiac arrest and many other cardiac diseases. Arrhythmia can be detected through a Holter monitor, event monitor, blood test, Electrocardiogram (ECG), etc. Here, ECG signals are used for automatic classification of heartbeat, which is required for arrhythmia detection. In the past, various machine learning approaches have been used but nowadays deep learning-based approaches are proposed mostly for getting better classification accuracy. In this chapter, a simplistic but robust customized deep learning model is implemented for automatic detection of arrhythmia. This model is performed on a standard dataset which is “MIT-BIH arrhythmia” dataset where an impressive classification accuracy of 99.74

Human Activity Recognition (HAR) is vital for healthcare, safety, smart homes, and transportation, enhancing well-being, security, and efficiency. It supports patient care, fitness tracking, workplace safety, and personalized automation while aiding sports performance, rehabilitation, and urban planning. This paper employs a pre-trained deep convolutional neural network, ResNet50V2 model, to classify daily activities of older people using heatmaps from accelerometer data collected from young and older adults. It examines daily physical activities in older adults ranging from fit to frail, comparing their performance with training data from the same age group and assessing models on individuals both with and without walking aids. The proposed ResNet-50V2 model achieved 98.2

In today’s context, the early detection of leaf diseases is paramount, saving considerable human effort and time. This paper addresses the concerned domain by utilizing some popular pre-trained deep learning models: DenseNet201, DenseNet121, ResNet50V2, ResNet101V2, VGG16, VGG19, MobileNetV1, and MobileNetV2. Subsequently, these models undergo training on the potato leaf disease and fruit infection disease datasets, with DenseNet121, MobileNetvV1, and ResNet101V2 identified as the most effective performers. For improving the detection performance, attention blocks are introduced into the models which significantly enhance their respective accuracy scores to 82

Over the past ten years, a lot of research has been done on automatic vehicle recognition utilizing machine learning approaches. The majority of earlier research on vehicle detection was done on datasets like as GTI, which mostly comprise well-organized road scenarios. These datasets do not accurately depict road scenes in areas with a lesser transportation infrastructure. As a result, the IDD dataset—which represents unstructured traveling situations—includes a large variety of vehicle classes, and shows adequate intra-class variability for vehicles—has been used to study vehicle recognition in this work. The suggested approach uses three neural networks—DenseNet21, SqueezeNet, and EfficientNet-B0—to extract features, which are then concatenated. An SVM classifier that uses a linear kernel has been trained using this fused collection of features. On a portion of the IDD dataset, the proposed methodology yields 89.73

Vehicle classification is important in many industries, such as surveillance, traffic flow monitoring, and automotive technology. Deep learning has given significant impetus to vehicle classification, but achieving high accuracy using this technique, especially in real-world applications, remains an elusive goal. The contribution of this work focuses on a fine-tuned EfficientNetB3 model for handling those issues. We tested our model on two benchmark datasets, such as PoribohonBD and JUTVCDv1, achieving as high as 98.70% and 97.37% classification accuracies, respectively, surpassing the previously best results of other methods. Further, pruning the model achieved not only improved performance but also the enhanced computation efficiency desired for real-time Vehicle Classification. The source code will be publicly available on https://github.com/Utathyaworks/Vehicle_classification

Brain tumors pose a singularly formidable threat in contemporary healthcare due to their diverse histological profiles and unpredictable clinical behavior. Their spectrum ranges from slow-growing benign tumors to highly aggressive malignancies in sensitive anatomical locations. This necessitates an intensified focus on their pathophysiology and demands precise characterization for patient-specific therapeutic solutions. Techniques to correctly identify brain tumors using artificial intelligence are often employed for addressing segmentation and detection tasks; however, the lack of generalizable results hinders medical practitioners from incorporating them into the diagnostic process. Predominantly reliant on Magnetic Resonance Imaging, research on other imaging methods like Positron Emission Tomography & Computed Tomography, is scarce due to a dearth of open-access datasets. Our study proposes a robust MBTC-Net framework by leveraging EfficientNetV2B0 for extracting high-dimensional feature maps, followed by reshaping into sequences and applying multi-head attention to capture contextual dependencies. After reintroducing the attention output into a spatial structure, we perform average pooling before transitioning to dense layers, enhanced with batch normalization and dropout. The model is fine-tuned with the Adamax optimizer to classify various kinds of brain tumors using softmax from T1-weighted, T1 Contrast-Enhanced, & T2-weighted MRI sequences and CT scans. To reduce the risk of overfitting, measures such as stratified 5-fold cross-validation have been extensively implemented across 3 open-access Kaggle datasets, obtaining 97.54 ​% (15-class), 97.97 ​% (6-class), and 99.34 ​% (2-class) accuracies, respectively. We have also applied Grad-CAM to decipher and visually analyze the predictions made by this framework. This research underscores the need for multimodal training of CT scans and MRI sequences for deploying a sturdy framework in real-time environments and advancing the well-being of patients.

Schizophrenia is a persistent and serious mental illness that leads to distortions in cognition, perception, emotions, speech, self-awareness, and actions. Affecting about 1% of people worldwide, schizophrenia usually emerges in late adolescence or early adulthood. It is characterized by symptoms like hallucinations, delusions, disorganized speech, and cognitive impairments. Despite significant research efforts, the exact cause of schizophrenia is still not fully understood, highlighting the need for continuous investigation into new diagnostic and treatment methods. The electroencephalogram (EEG), which measures brain electrical activity using scalp electrodes, is crucial in schizophrenia research due to its ability to detect subtle brain activity changes due to high temporal information and provide valuable insights into brain function. Many methods have been proposed to identify schizophrenia for diagnosis. Different machine learning and deep learning models have been used to improve the detection of schizophrenia. Through transfer learning using deep learning models, relevant features are selected automatically, outperforming traditional methods in accuracy and speed. Our paper introduces a three-stage framework for detection of schizophrenia from EEG signals. An image encoding method has been used to encode EEG signals to scalogram images to get both spatial and temporal information of the time series data. Using these images in the second step, two pre-trained deep learning models are implemented using transfer learning to extract features for the detection of schizophrenia. In the third step, a newly developed Average subtraction wrapper-based feature selection method has been proposed to lower the number of irrelevant features. The proposed framework has been tested on two datasets. The first (M.S.U) dataset is from M.V. Lomonosov Moscow State University which contains EEG data of 84 individuals where 45 individuals are with schizophrenia symptoms and the rest are 39 individuals are healthy. The second RepOD dataset contains EEG data of 28 individuals where both schizophrenic and healthy individuals are equal in number. Our framework achieved 99.67% and 99.97% accuracy on the first and second dataset, respectively. On both the datasets, our proposed framework outperformed state of the art results.

This paper presents a novel methodology for depression detection using Electroencephalogram (EEG) signals from the Multi-modal Open Dataset for Mental Disorder Analysis (MODMA). The proposed approach integrates advanced signal processing techniques with a Recurrent Neural Network (RNN) classification model to objectively identify depressive states. The methodology consists of three key stages: data preprocessing, feature extraction, and RNN-based classification. EEG data is preprocessed using bandpass filtering, reshaping, and splitting into training and testing sets, preparing it for deep learning model input. The RNN-based model is trained on these features, achieving an accuracy of 77

This paper presents a thorough evaluation of machine learning algorithms for assessing the risk of having diabetes with the help of the Pima Indian Diabetes dataset. In view of the global diabetes epidemic, timely and precise risk assessment is imperative. Our study involves an in-depth exploration of the data, uncovering a robust correlation between glucose levels and the likelihood of diabetes. We deploy a diverse set of nine machine learning models, encompassing logistic regression, decision trees, random forests, AdaBoost, support vector machine, K-nearest neighbors, Naive Bayes, XGBoost, and an artificial neural network approach. The results illuminate the strengths and weaknesses of each model, providing valuable insights for potential clinical applications. The logistic regression approach, in particular, showcases its ability to capture intricate patterns within the dataset, underscoring its effectiveness in diabetes risk assessment. In addition to traditional classifiers, we introduce an ensemble model that combines the strengths of five best performing classifiers. These findings not only improve the accuracy of diabetes risk assessment but also establish a benchmark for future research in medical diagnosis. Identifying the most effective model assists healthcare practitioners in early intervention and tailored treatment strategies. This research advances the field of healthcare analytics, facilitating more informed decisions in diabetes prevention and management. The integration of a robust ensemble model approach broadens the scope of potential applications, marking a significant contribution to the field.

Vehicle make and model recognition (VMMR) using still images is a challenging research problem. Automatic VMMR systems have many real-life applications that include surveillance. In this paper, initially, we have used five standard convolutional neural network (CNN) models, namely Inceptionv3, Xception, InceptionResNetv2, MobileNetV2, and ResNet152v2 for VMMR. We have also used an attention mechanism to these models. To increase accuracy of the overall model, we have chosen three best base learners from these five CNN models, and formed an ensemble model. The final model is called XMR_Net, where X stands for Xception, M stands for MobileNet, and R stands for ResNet152v2. For experimental evaluation, we have used two benchmark datasets, a recently published dataset called Vehicle Images dataset and VMMRdb-53 dataset. We have achieved satisfactory outcomes with accuracy scores of 95 https://github.com/JUVCSE/XMRNET .

Cervical cancer, a disease that can be prevented with early detection, remains a global health concern, mainly in the remote and underrepresented regions of the world. Our study introduces an ensemble method combining Swin Transformer, Vision Transformer (ViT), and ResNet50 to classify cancerous and precancerous cells in Pap smear images. Each model has almost unique characteristics—hierarchical vision processing, global attention mechanisms, and robust feature extraction—the ensemble uses Softmax-weighted averaging to combine the classification power of each model and reduce false positives. Our ensemble model is evaluated on the SipakMed dataset of 4,049 images and 966 manually cropped cell clusters, achieving an impressive accuracy of 97.87

Background: Parkinson’s disease (PD) is one of the most prevalent, widespread, and intricate neurodegenerative disorders. According to the experts, at least 1% of people over the age of 60 are affected worldwide. In the present time, the early detection of PD remains difficult due to the absence of a clear consensus on its brain characterization. Therefore, there is an urgent need for a more reliable and efficient technique for early detection of PD. Using the potential of electroencephalogram (EEG) signals, this study introduces an innovative method for the detection or classification of PD patients through machine learning, as well as a more accurate deep learning approach. Methods: We propose an innovative EEG-based PD detection approach by integrating advanced spectral feature engineering with machine learning and deep learning models. Using (a) the UC San Diego Resting State EEG dataset and (b) IOWA dataset, we extract a standardized EEG feature from five key frequency bands—alpha, beta, theta, gamma, delta (α,β,θ,γ,δ) and employ an SVM (Support Vector Machine) classifier as a baseline, achieving a notable accuracy. Furthermore, we implement a deep learning classifier (CNN) with a complex multi-dimensional feature set by combining power values from all frequency bands, which gives superior performance in distinguishing PD patients (both with medication and without medication states) from healthy patients. Results: With the five-fold cross-validation on these two datasets, our approaches successfully achieve promising results in a subject dependent scenario. The SVM classifier achieves competitive accuracies of 82% and 94% in the UC San Diego Resting State EEG dataset (using gamma band) and IOWA dataset, respectively in distinguishing PD patients from non-PD patients in subject. With the CNN classifier, our model is able to capture major cross-frequency dependencies of EEG; therefore, the classification accuracies reach beyond 96% and 99% with those two datasets, respectively. We also perform our experiments in a subject independent environment, where the SVM generates 68.09% accuracy. Conclusions: Our findings, coupled with advanced feature extraction and deep learning, have the potential to provide a non-invasive, efficient, and reliable approach for diagnosing PD, with further work aimed at enhancing feature sets, inclusion of a large number of subjects, and improving model generalizability across more diverse environments.