Research on Mathematical Method Image Classification of Convolutional Neural Network Based on Firework Algorithm Optimization

• Published in: Wireless communications and mobile computing Vol. 2022; no. 1
• Main Authors: Cui, Liping, Yang, Xin, Liu, Yan
• Format: Journal Article
• Language: English
• Published: Oxford Hindawi 2022; John Wiley & Sons, Inc
• Subjects: Algorithms; Artificial neural networks; Brain research; Classification; Datasets; Fireworks; Image classification; Neural networks; Object recognition; Optimization; Performance enhancement; Performance indices; Recurrent neural networks; Search engines
• ISSN: 1530-8669; 1530-8677; 1530-8677
• DOI: 10.1155/2022/8646994

The exhibition of famous convolutional brain organizations (CNNs) for distinguishing objects progressively video takes care of is inspected in this exploration. AlexNet, GoogLeNet, and ResNet50 are the most well-known convolutional neural networks for object discovery and item classification arrangement from pictures. To survey the exhibition of various kinds, a variety of photo informative indexes are provided by CNNs. Standard benchmark datasets for estimating a convolutional neural organization’s exhibition include ImageNet, CIFAR10, CIFAR100, and MNIST picture informational indexes. The performance of the three well-known channels, Alexandra cash flow, search engine net, and recurrent neural networks, is investigated in this research. Because analyzing a cable network efficiency on a single dataset does not demonstrate all of its possibilities and limits, we mentioned two of the most prominent large datasets for research: significantly improve performance, FARCICAL, and CIFAR110. Clips are exploited as testing statistics rather than teaching statistics; it should have been mentioned. GoogLeNet and ResNet50, in comparison to AlexNet, are better at recognizing objects with greater precision. Furthermore, the performance of trained CNNs varies significantly across different object categories, and we will analyze the possible causes for this. The characterization rate is the goal work assessed by PSO in the main methodology; in the subsequent methodology, the fireworks produce various boundaries per layer, and the goal work is made out of the recognition rate related to the Akaike data model, which assists with finding the best organization per layer. As per the discoveries, the proposed strategy delivered positive results with a recognition pace of more prominent than close to 100%, exhibiting serious outcomes when contrasted with other cutting edge draws near.