WeedDetectNet: A Novel Deep Learning Framework for Weed Detection in Cassava Crops

Abstract

Background: Weed detection is an integral aspect of precision agriculture and the detection of multiple weed types in fields presents a challenge in cassava production because the presence of different weed types may hinder the ability to analyze the crop accurately. The capability to detect weeds in field pictures correctly is crucial in ensuring that the intelligent agricultural system developed is able to distinguish between crops and weed plantations. Methods: This study presents a deep learning-based framework for automated weed detection in cassava fields. The proposed customized convolutional neural network (CCNN) accurately distinguishes cassava plants from different weed categories under varying field conditions, demonstrating its effectiveness for agricultural image analysis and precision agriculture research. The system is trained to segment the images into cassava plants, broadleaf weeds, grassy weeds and sedges. Result: Our experiments demonstrate that the proposed system detects images with an accuracy of 99.56%, outperforming the state-of-the-art VGG16 model based on hand-designed features. The experimental results demonstrate that the proposed CCNN model achieves highly accurate weed detection and classification under diverse field conditions.