Title: Improvement of convolutional neural networks in image classification and recognition

Authors: Pengju Xia

Addresses: School of Finance and Economics, Xuchang Vocational Technical College, Xuchang, 461000, China

Abstract: Image classification, a vital deep learning technique in vision, has been widely applied in fields such as agricultural detection. However, traditional methods often suffer from low efficiency and weak analytical capability. To address this, we propose a CNN-ELM based image classification model incorporating a feature reorganisation attention mechanism and capsule networks for extracting complex semantic information. Experimental results show that when the iteration number of the fusion algorithm reaches 200, the average loss value is 0.026, significantly lower than that of the transfer learning ensemble algorithm (0.082) and the FPGA algorithm (0.087). Moreover, the hybrid model achieves a recognition accuracy of 0.99 for animal-type images, compared to 0.91 with the transfer learning ensemble model. These findings demonstrate that the proposed approach effectively enhances feature recognition and information capture, offering promising potential for structural information analysis and advancing medical imaging applications.

Keywords: convolutional neural network; CNN; image classification and recognition; feature reorganisation attention mechanism; capsule network; extreme learning machine; ELM.

DOI: 10.1504/IJICA.2025.149777

International Journal of Innovative Computing and Applications, 2025 Vol.15 No.4, pp.256 - 270

Received: 09 Jun 2025
Accepted: 18 Aug 2025
Published online: 12 Nov 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article