Images taken with different sensors and transmitted through different channels can be noisy. In such conditions, the image most often suffers from random-valued impulse noise. Denoising an image is an important part of image preprocessing before recognition by a neural network. The accuracy of image recognition by a neural network directly depends on the intensity of image noise. This paper presents a three-stage image cleaning and recognition system, which includes a developed detector of pulsed noisy pixels, a filter for cleaning found noisy pixels based on an adaptive median, and a neural network program for recognizing cleaned images. It was noted that at low noise intensities, cleaning is practically not required, but noise with an intensity of more than 10% can seriously damage the image and reduce recognition accuracy. As a training base for noise, cleaning, and recognition, the CIFAR10 digital image database was used, consisting of 60,000 images belonging to 10 classes. The results show that the proposed neural network recognition system for images affected by to random-valued impulse noise effectively finds and corrects damaged pixels. This helped to increase the accuracy of image recognition compared to existing methods for cleaning random-valued impulse noise.

In this paper we propose the fractional gradient descent for increasing the training and work of modern neural networks. This optimizer searches the global minimum of the loss function considering the fractional gradient directions achieved by Riemann-Liouville, Caputo, and Grunwald-Letnikov derivatives. The adjusting of size and direction of the fractional gradient, supported by momentum and Nesterov condition, let the proposed optimizer descend into the global minimum of loss functions of neural networks. Utilizing the proposed optimization algorithm in a linear neural network and a visual transformer lets us attain higher accuracy, precision, recall, Macro F1 score by 1.8-4 percentage points than known analogs than state-of-the-art methods in solving pattern recognition problems on images from MNIST and CIFAR10 datasets. Further research of fractional calculus in modern neural network methodology can improve the quality of solving various challenges such as pattern recognition, time series forecasting, moving object detection, and data generation.

Food shortages are becoming increasingly urgent due to the growing global population. Enhancing oil crop yields, particularly sunflowers, is key to ensuring food security and the sustainable provision of vegetable fats essential for human nutrition and animal feed. However, sunflower yields are often reduced by diseases, pests, and other factors. Remote sensing technologies, such as unmanned aerial vehicle (UAV) scans and satellite monitoring, combined with machine learning algorithms, provide powerful tools for monitoring crop health, diagnosing diseases, mapping fields, and forecasting yields. These technologies enhance agricultural efficiency and reduce environmental impact, supporting sustainable development in agriculture. This systematic review aims to assess the accuracy of various machine learning technologies, including classification and segmentation algorithms, convolutional neural networks, random forests, and support vector machines. These methods are applied to monitor sunflower crop conditions, diagnose diseases, and forecast yields. It provides a comprehensive analysis of current methods and their potential for precision farming applications. The review also discusses future research directions, including the development of automated systems for crop monitoring and disease diagnostics.