Modified xLSTM for Compression and Decompression of Multimodal Agricultural Data in Low-Resource Settings

Abstract

Efficient compression techniques are essential for handling large datasets, especially in low-resource agricultural settings where bandwidth and storage are limited. This paper introduces a novel approach that combines a modified extended Long Short-Term Memory (xLSTM) network with multiplicative LSTM (mLSTM) cells for compressing and decompressing multimodal agricultural data. The key contribution lies in tailoring the xLSTM-mLSTM architecture specifically for agricultural data, capturing unique patterns to enhance compression efficiency. Specifically, we focus on agricultural datasets comprising textual labels and images. The proposed method combines xLSTM with mLSTM cells for text data compression and employs a convolutional autoencoder for image data. We compare our approach with existing compression methods applied to agricultural data, demonstrating superior performance in terms of compression ratio and reconstruction quality. Experimental results demonstrate the effectiveness of the approach, achieving significant data size reduction while maintaining acceptable reconstruction quality, as evidenced by metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM).