KonvLiNA: Integrating Kolmogorov-Arnold Network with Linear Nystr\"om Attention for feature fusion in Crop Field Detection

TL;DR

KonvLiNA is a novel framework combining convolutional Kolmogorov-Arnold Networks and Nyström attention to improve crop field detection accuracy, outperforming existing methods in agricultural image analysis.

Contribution

The paper introduces a hybrid model integrating KAN and Nyström attention, enhancing feature extraction for crop detection in complex environments.

Findings

Achieved 0.415 AP and 0.459 AR on rice dataset with Swin-L backbone.

Outperformed YOLOv8 significantly in crop field detection tasks.

Demonstrated competitive performance on COCO dataset across object sizes.

Abstract

Crop field detection is a critical component of precision agriculture, essential for optimizing resource allocation and enhancing agricultural productivity. This study introduces KonvLiNA, a novel framework that integrates Convolutional Kolmogorov-Arnold Networks (cKAN) with Nystr\"om attention mechanisms for effective crop field detection. Leveraging KAN adaptive activation functions and the efficiency of Nystr\"om attention in handling largescale data, KonvLiNA significantly enhances feature extraction, enabling the model to capture intricate patterns in complex agricultural environments. Experimental results on rice crop dataset demonstrate KonvLiNA superiority over state-of-the-art methods, achieving a 0.415 AP and 0.459 AR with the Swin-L backbone, outperforming traditional YOLOv8 by significant margins. Additionally, evaluation on the COCO dataset showcases competitive performance across small, medium, and large objects, highlighting KonvLiNA efficacy in diverse agricultural settings. This work highlights the potential of hybrid KAN and attention mechanisms for advancing precision agriculture through improved crop field detection and management.