Abstract:Kernel-based statistical methods are efficient, but their performance depends heavily on the selection of kernel parameters. In literature, the optimization studies on kernel-based chemometric methods is limited and often reduced to grid searching. Previously, the authors introduced Kernel Flows (KF) to learn kernel parameters for Kernel Partial Least-Squares (K-PLS) regression. KF is easy to implement and helps minimize overfitting. In cases of high collinearity between spectra and biogeophysical quantities in spectroscopy, simpler methods like Principal Component Regression (PCR) may be more suitable. In this study, we propose a new KF-type approach to optimize Kernel Principal Component Regression (K-PCR) and test it alongside KF-PLS. Both methods are benchmarked against non-linear regression techniques using two hyperspectral remote sensing datasets.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to optimize the parameter selection of kernel - based methods in hyperspectral image data. Specifically, the paper proposes a new Kernel Flows (KF) - based method to optimize the kernel parameters in Kernel Principal Component Regression (K - PCR) and compares it with the optimized Kernel Partial Least Squares (K - PLS). The paper benchmarks these two methods using two hyperspectral remote sensing datasets to evaluate their performance in nonlinear regression techniques. ### Background Hyperspectral imaging technology shows great potential in environmental monitoring, such as in water quality monitoring, agriculture, vegetation health, and climate observation. However, although traditional deep - learning methods have been successful in creating retrieval models from hyperspectral data, they also have problems such as over - fitting, poor interpretability, and the need for large amounts of data. In contrast, traditional chemometric models (such as Partial Least Squares regression PLS) require a smaller amount of training data, but have a weaker ability to model nonlinear relationships. Therefore, kernelized chemometric methods can better handle nonlinear dependencies, but the learning of kernel functions and their parameters is a complex problem. ### Research Objectives 1. **Optimize Kernel Parameters**: Use the Kernel Flows (KF) method to optimize the kernel parameters in Kernel Principal Component Regression (K - PCR). 2. **Performance Evaluation**: Through two hyperspectral remote - sensing datasets, compare the performance of the optimized K - PCR and K - PLS with nonlinear regression techniques. 3. **Improve Loss Function**: Propose a new loss function to improve the performance of KF when optimizing K - PCR and K - PLS. ### Methods - **Kernel Flows (KF)**: KF is a method that minimizes the loss function through cross - validation and Stochastic Gradient Descent (SGD) and can effectively learn kernel parameters. - **K - PCR and K - PLS**: The paper extends previous work on K - PLS, applies it to K - PCR, and proposes a new loss function to adapt to the characteristics of K - PCR. - **Datasets**: Two hyperspectral remote - sensing datasets, one is a soil moisture soft sensor, and the other is a vegetation feature model. ### Results - **Soil Moisture Prediction**: The optimized K - PCR model shows higher accuracy in soil moisture prediction, especially on the test set. - **Plant Nitrogen Content Prediction**: In the vegetation feature model, the optimized K - PLS performs best, followed by Gaussian Process Regression (GPR). K - PLS and K - PCR are superior to linear methods and, due to their fewer parameters and cross - validation nature, reduce the risk of over - fitting. ### Discussion - **Performance Improvement**: The new loss function brings a 33% performance improvement in K - PCR and a 2% performance improvement in K - PLS. - **Applicability**: This method is applicable not only to soil moisture prediction but also to the prediction of other biogeophysical quantities, such as plant nitrogen content. - **Future Work**: Future research can explore more effective sample selection strategies to optimize the construction of the kernel matrix and training time for large - scale datasets. ### Conclusion The paper proposes an optimization method based on Kernel Flows to optimize the kernel parameters in K - PCR. The experimental results show that this method performs well in both soil moisture prediction and plant nitrogen content prediction and has broad application prospects.

Kernel-based retrieval models for hyperspectral image data optimized with Kernel Flows

Cross-validatory Framework for Optimal Parameter Estimation of KPCA and KPLS Models

A Kernel Function Optimization And Selection Algorithm Based On Cost Function Maximization

A Kernel Spatial Complexity-Based Nonlinear Unmixing Method of Hyperspectral Imagery.

A Spectral-Texture Kernel-Based Classification Method for Hyperspectral Images.

Hyperspectral Unmixing Based on Incremental Kernel Nonnegative Matrix Factorization.

Kernel-Based Nonparametric Fisher Classifier for Hyperspectral Remote Sensing Imagery

Efficient Photometric Stereo Using Kernel Regression

Randomized kernels for large scale Earth observation applications

Kernel Low-Rank Entropic Component Analysis for Hyperspectral Image Classification

Hyperspectral Imaging Target Detection Based on Improved Kernel Principal Component Analysis

Retrieval of Physical Parameters With Deep Structured Kernel Regression

Kernel Supervised Ensemble Classifier for the Classification of Hyperspectral Data Using Few Labeled Samples

Kernel-based partial least squares: application to fingerprint-based QSAR with model visualization

Optimized Kernel Minimum Noise Fraction Transformation for Hyperspectral Image Classification.

A New Convolutional Kernel Classifier for Hyperspectral Image Classification

Kernel partial diagnostic robust potential to handle high-dimensional and irregular data space on near infrared spectral data

A support vector machine classifier based on a new kernel function model for hyperspectral data

A Kernel Optimization Algorithm Based on Fusion Kernel for High-Resolution Range Profiles Recognition

Nonlinear Unmixing for Hyperspectral Images via Kernel-Transformed Bilinear Mixing Models

Use of customizing kernel sparse representation for hyperspectral image classification.