Abstract:In order to meet the increasing demand for crops under challenging climate conditions, efficient and sustainable cultivation strategies are becoming essential in agriculture. Targeted herbicide use reduces environmental pollution and effectively controls weeds as a major cause of yield reduction. The key requirement is a reliable weed detection system that is accessible to a wide range of end users. This research paper introduces a self-built, low-cost, multispectral camera system and evaluates it against the high-end MicaSense Altum system. Pixel-based weed and crop classification was performed on UAV datasets collected with both sensors in maize using a U-Net. The training and testing data were generated via an index-based thresholding approach followed by annotation. As a result, the F1-score for the weed class reached 82% on the Altum system and 76% on the low-cost system, with recall values of 75% and 68%, respectively. Misclassifications occurred on the low-cost system images for small weeds and overlaps, with minor oversegmentation. However, with a precision of 90%, the results show great potential for application in automated weed control. The proposed system thereby enables sustainable precision farming for the general public. In future research, its spectral properties, as well as its use on different crops with real-time on-board processing, should be further investigated.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to realize an efficient and sustainable weed detection system in agriculture. Specifically, the research aims to develop a UAV - based weed detection method using a low - cost multispectral camera system and compare its performance with that of a high - end multispectral camera system. By using deep - learning techniques (especially the U - Net architecture), the researchers hope to evaluate the feasibility and accuracy of the low - cost system in weed detection, so as to provide a more economical and more popular solution. ### Research Background and Motivation With the rapid growth of the global population, the reduction of arable land and the unreliable agricultural production conditions brought by climate change, agriculture is under increasing pressure. In order to meet the future's growing demand for food, innovative solutions are needed to improve agricultural efficiency and make these solutions widely available. Weeds are the main cause of crop yield reduction because they compete with crops for nutrients, water, sunlight and space. Although the use of herbicides is the most common and effective method to control weeds, excessive use will lead to irreversible ecological damage, such as groundwater pollution, soil pollution and biodiversity loss. Therefore, site - specific weed control in precision agriculture can significantly reduce the use of herbicides, thus achieving sustainable crop management. ### Research Objectives 1. **Hardware and Software Development**: Introduce the hardware and software development of a low - cost multispectral camera system (LCS) for agricultural weed detection. 2. **Performance Comparison**: In practical applications, compare the proposed LCS with the high - performance MicaSense Altum multispectral camera system and evaluate its classification accuracy in weed detection. 3. **Weakness Identification and Discussion**: Identify the potential weaknesses of LCS, determine its applicable scenarios and make suggestions for the further development of the system. ### Methods 1. **Data Collection**: - Conduct UAV flights in a cornfield in Lower Saxony, Germany, using MicaSense Altum and a self - built low - cost multispectral camera system (LCS). - The flight altitude is 12 meters to ensure the visibility of weeds in the early vegetation stage. 2. **Data Pre - processing**: - Perform geometric transformation and cropping on images of different channels to eliminate distortion and overlap. - Calculate the NDVI index, generate a vegetation mask and perform manual annotation. 3. **Semantic Segmentation**: - Use the U - Net architecture for pixel - level classification, and the input data include red, green and near - infrared (NIR) bands. 4. **Evaluation Metrics**: - Use metrics such as Jaccard index, precision, recall and F1 - score to evaluate the performance of the model. ### Results 1. **Qualitative Results**: - The accuracy of training and test data was evaluated by manually correcting the vegetation mask. The results show that LCS data has slight over - segmentation and imprecise boundary problems, especially in the detection of small leaves and weeds. 2. **Quantitative Results**: - The Jaccard score of U - Net on the Altum dataset reaches 96.75%, while it is 93.27% on the LCS dataset. - The detection precision of the soil category is very high, almost reaching 100%. - The detection precision of the crop category on the LCS dataset is slightly lower, but the recall rate is still high. - The detection precision of the weed category on the LCS dataset is significantly lower, especially in the detection of small weeds and overlapping areas. ### Conclusions The research results show that the low - cost multispectral camera system has great potential in weed detection, although it does not perform as well as high - end systems in some details. Through further optimization and improvement, this low - cost system is expected to become an affordable and effective tool in precision agriculture.

Weed Detection from Unmanned Aerial Vehicle Imagery Using Deep Learning—A Comparison between High-End and Low-Cost Multispectral Sensors

Machine learning approaches to automate weed detection by UAV based sensors

Estimating Canopy Surface Height of Wheat and Corn Crops in Reclaimed Cropland Using Multispectral Images from a Small Unmanned Aircraft System

Deep Learning-Based Object Detection System for Identifying Weeds Using UAS Imagery

Fusion of Pixel and Object-Based Features for Weed Mapping Using Unmanned Aerial Vehicle Imagery

WeedMap: A Large-Scale Semantic Weed Mapping Framework Using Aerial Multispectral Imaging and Deep Neural Network for Precision Farming

Machine Learning Interventions for Weed Detection using Multispectral Imagery and Unmanned Aerial Vehicles -- A Systematic Review

Cognitive Computing Advancements: Improving Precision Crop Protection through UAV Imagery for Targeted Weed Monitoring

Weed Mapping with UAS Imagery and a Bag of Visual Words Based Image Classifier

Comparison of Object Detection and Patch-Based Classification Deep Learning Models on Mid- to Late-Season Weed Detection in UAV Imagery

UAV Low-Altitude Remote Sensing for Precision Weed Management

Deep Learning with Unsupervised Data Labeling for Weeds Detection on UAV Images

Deep learning-based identification system of weeds and crops in strawberry and pea fields for a precision agriculture sprayer

UAV-Assisted Thermal Infrared and Multispectral Imaging of Weed Canopies for Glyphosate Resistance Detection

Optimized Deep Learning Model as a Basis for Fast UAV Mapping of Weed Species in Winter Wheat Crops

Harnessing UAVs and deep learning for accurate grass weed detection in wheat fields: a study on biomass and yield implications

Advancements of UAV and Deep Learning Technologies for Weed Management in Farmland

Weed Warden: A low-cost weed detection device implemented with spectral triad sensor for agricultural applications

Transferring learned patterns from ground-based field imagery to predict UAV-based imagery for crop and weed semantic segmentation in precision crop farming

weedNet: Dense Semantic Weed Classification Using Multispectral Images and MAV for Smart Farming