KonvLiNA: Integrating Kolmogorov-Arnold Network with Linear Nyström Attention for feature fusion in Crop Field Detection
This work addresses crop field detection for precision agriculture, offering a novel hybrid method with strong specific gains.
The paper tackles crop field detection by introducing KonvLiNA, a framework integrating Convolutional Kolmogorov-Arnold Networks with Nyström attention, which achieved a 0.415 AP and 0.459 AR on a rice crop dataset, outperforming YOLOv8.
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 Nyström attention mechanisms for effective crop field detection. Leveraging KAN adaptive activation functions and the efficiency of Nyström 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.