Invasive weed optimization with deep transfer learning for multispectral image classification model

Multispectral image classification is a field of static learning with non-stationary input data assumptions. The evolution of Industry 4.0 has resulted in the development of multispectral images in several application areas. The classification of multispectral images is a tedious process due to its complex characteristics of including spectral as well as spatial features. The recent developments of machine learning (ML) and deep learning (DL) paves a way for the design of effectual computer vision tasks such as object recognition, classification, etc. With this motivation, this paper presents an Invasive Weed Optimization with Capsule Network for Multispectral Image Classification (IWOCN-MSIC) technique. The major intention of the IWOCN-MSIC technique incorporates multilevel discrete wavelet transform (DWT) based image decomposition technique, which decomposes the input image into distinct subbands. Besides, IWO algorithm with Capsule Network (CapsNet) model is applied to derive a useful subset of feature vectors. The complex computation in Capsule Networks(CapsNet) requires hyper-parameters to achieve high classification outputs, which requires more computational time and effort. To overcome this difficulty, a bio-inspired meta-heuristic strategy based Invasive Weed optimization is proposed in this research paper. It allows one to automatically search for and select the appropriate values of CapsNet hyper-parameters. Finally, Deep Denoising Auto-Encoder (DDAE) model is employed for the detection and classification of multispectral images. The experimental result analysis of the IWOCN-MSIC technique reported the better performance of the IWOCN-MSIC technique over the recent state of art approaches.