deep learning

Development of a deep learning-based system in Python 3.9 with YOLOv5: A case study on real-time fish counting based on classification

This study developed a real-time fish classification and counting system for six types of fish using the YOLOv5 machine learning model with high accuracy.  The system achieved an F1-score of 0.87 and a precision confidence curve with an all-classes value of 1.00 at a confidence level of 0.920, demonstrating the model's reliability in object detection and classification.  Real-time testing showed that the system could operate quickly and accurately under various environmental conditions with an average inference speed of 30 FPS.  However, several challenges remain, such

Forecasting solar energy generation using deep learning models

The application of deep learning models for forecasting solar energy generation is considered.  An analysis and comparison of the efficiency of recurrent (LSTM, GRU), convolutional (CNN), and temporal convolutional networks (TCN) for forecasting time series of solar energy generation were conducted.  The possibility of improving forecasting accuracy by constructing a hybrid model combining ARIMA and CNN was explored.  The results of experiments for different EU countries are presented, and a comparison of models in terms of forecasting accuracy and computational efficiency is performed as w

IDENTIFYING GRAPE DISEASES BY IMAGES USING ARTIFICIAL INTELLIGENCE METHODS

The paper uses modern artificial intelligence methods to investigate models and methods for determining grape disease. The existing methodologies for classification and recognition by images of grape diseases using neural networks are analyzed. Several problems for improving recognition results are highlighted.

ACTION-MASKED REINFORCEMENT LEARNING TECHNOLOGY FOR ORDER SCHEDULING

The problem of high-performance and efficient order scheduling is a common combinatorial optimization problem in various industrial contexts. Creation of a model capable of generating schedules balanced in terms of quality and computational time poses a significant challenge due to the large action space. This study proposes a high-performant environment and a reinforcement learning model for allocating orders to resources using a mechanism of invalid action masking.

Numerical simulation by Deep Learning for discrete nonlinear problems involving the anisotropic p(.)-Laplacian

In this paper, we establish the existence of a class of discrete nonlinear systems involving the anisotropic $\vec{p}(\cdot)$-Laplacian operator using an optimization based approach.  We then simulate the solutions by implementing a deep learning model.  The numerical results demonstrate that the proposed method is stable and robust compared to conventional approaches such as the Newton–Krylov method.

Structure of the Information System for Predicting and Interpreting Changes in the State of the Service User

The paper investigates the problem of predicting changes in user states (including churn) based on session data using deep neural networks. The paper considers the use of long short-term memory models and convolutional neural networks, as well as the use of byte pair coding for data pre-processing. The functionality of the developed information system for forecasting changes in the state of users and interpreting forecasting models, which combines methods of data analysis, building forecasting models and explaining the results, is analysed.

Real-time Anomaly Detection in Distributed Iot Systems:a Comprehensive Review and Comparative Analysis

The rapid expansion of the Internet of Things (IoT) has resulted in a substantial increase of diverse data from distributed devices. This extensive data stream makes it increasingly important to implement robust and efficient real-time anomaly detection techniques that can promptly alert about issues before they could escalate into critical system failures.

Evaluation of Deep Learning-based Super-resolution Methods for Enhanced Facial Identification Accuracy

This paper presents a comparative analysis of modern super-resolution (SR) methods for improving the accuracy of face recognition in video surveillance systems. The low quality of images obtained from surveillance cameras is a significant obstacle to effective person identification, making the use of SR methods particularly relevant.

ADAPTIVE OBJECT RECOGNITION THROUGH A META-LEARNING APPROACH FOR DYNAMIC ENVIRONMENTS

Object recognition systems often struggle to maintain accuracy in dynamic environments due to challenges such as lighting variations, occlusions, and limited training data. Traditional convolutional neural networks (CNNs) require extensive labeled datasets and lack adaptability when exposed to new conditions. This study aims to develop an adaptive object recognition framework that enhances model generalization and rapid adaptation in changing environments.

CREATING A FACIAL RECOGNITION SYSTEM BASED ON A NEURAL NETWORK THAT WORKS IN REAL TIME

The article is devoted to the theoretical foundations and practical aspects of facial recognition using neural networks. The paper analyzes various approaches to feature extraction from facial images, as well as methods for training neural networks for recognition. Special attention is paid to problems related to variations in lighting, facial expressions, and other factors affecting recog- nition accuracy. The paper describes the creation and optimization of a neural network for real-time human face recognition.