Low-Resource Crop Classification from Multi-Spectral Time Series Using Lossless Compressors

TL;DR

This paper tackles crop classification from multispectral time series in low-resource settings where deep learning models struggle due to limited labeled data and high training costs. It introduces a non-training framework built around a Symbolic Representation Module, cross-transformations to create symbolic embeddings, and Multi-scale Normalised Compression Distance (MNCD) to quantify similarity, with classification performed by a simple $k$NN. The approach leverages lossless compressors to capture regularities in symbolic data, achieving competitive performance with large-scale DL models across three benchmarks and showing strong few-shot results. Its lightweight, training-free nature and robustness in real-world crop mapping make it a practical alternative for resource-constrained agricultural monitoring.

Abstract

Deep learning has significantly improved the accuracy of crop classification using multispectral temporal data. However, these models have complex structures with numerous parameters, requiring large amounts of data and costly training. In low-resource situations with fewer labeled samples, deep learning models perform poorly due to insufficient data. Conversely, compressors are data-type agnostic, and non-parametric methods do not bring underlying assumptions. Inspired by this insight, we propose a non-training alternative to deep learning models, aiming to address these situations. Specifically, the Symbolic Representation Module is proposed to convert the reflectivity into symbolic representations. The symbolic representations are then cross-transformed in both the channel and time dimensions to generate symbolic embeddings. Next, the Multi-scale Normalised Compression Distance (MNCD) is designed to measure the correlation between any two symbolic embeddings. Finally, based on the MNCDs, high quality crop classification can be achieved using only a k-nearest-neighbor classifier kNN. The entire framework is ready-to-use and lightweight. Without any training, it outperformed, on average, 7 advanced deep learning models trained at scale on three benchmark datasets. It also outperforms more than half of these models in the few-shot setting with sparse crop labels. Therefore, the high performance and robustness of our non-training framework makes it truly applicable to real-world crop mapping. Codes are available at: https://github.com/qinfengsama/Compressor-Based-Crop-Mapping.

Figures (6)

• Figure 1: These time series are discretized by using predetermined breakpoints $B$ to map the reflectivity to symbols. In the example above, with $l = 4$, the time series of band 1 is mapped to cdcdcbc and the time series of band 2 is mapped to bababaa.
• Figure 2: An illustrative description of cross-transforming in the time and channel dimensions to obtain multi-scale NCD $d_{p q}$ between pixels $S_p$ and $S_q$ after symbolic representation.
• Figure 3: Classification maps obtained from different models on a typical plot.
• Figure 4: Confusion matrix for 50-shot crop classification results on the T31TFM-1618 dataset.
• Figure 5: Results on three datasets for different alphabet lengths.