Review of Deep Learning Models Based on Hyperspectral Image Classification Techniques in Monitoring Crop Production

the importance of evaluating and ensuring the quality of agricultural and horticultural products has increased in recent years. Conventional techniques that require Field measurements, research, as well as statistics analysis are expensive, and time taking labor-intensive. As a novel technology, HSI has been very popular, especially because of its potential uses in the field of remote sensing, particularly in agriculture. But, there are a number of difficulties in identifying HSI data, including the complex non-linear relationship between spectral bands and spatial coordinates, limited availability of training data and the massive duplication of spectral bands. Interestingly, Deep Learning (DL) techniques have demonstrated remarkable efficacy in a number of HSI analytic applications, including agricultural ones. There is an urgent need for a thorough survey that will help academics understand the significant progress made and the fascinating potential avenues for future research in this area, given the growing interest in using DL approaches to leverage HSI data for agricultural applications. This study of the literature carefully gathers, examines, and talks about recent attempts to use DL approaches. Autoencoders (AE), Recurrent Neural Networks (RNN), Deep Belief Networks (DBN), Generative Adversarial Networks (GAN), Convolutional Neural Networks (CNN) (in 1D, 2D, and 3D configurations), Transfer Learning (TL), Semi-Supervised Learning (SSL), Few-Shot Learning (FSL), and Active Learning (AL) are just a few of the techniques that fall under this broad category. These methods are designed to tackle the particular difficulties presented by HSI analysis in agriculture. The performance demonstrated by these various approaches is assessed and discussed in this review. In light of this, the effectiveness of these methods has been thoroughly examined and debated in light of the findings of cutting-edge research on well-known land cover datasets repository on Github. The viability, adaptability, and affordability of using hyperspectral imagery (HSI) for crop classification in agricultural regions have been extensively established.