The comparison of pixel-based image analysis for detection of weeds in winter wheat from UAV imagery

Creating weed maps directly by growers is becoming increasingly common. In this study, an unmanned aerial vehicle (UAV) imaged a field infested by field thistle (Cirsium arvense). This paper compares four detection methods that can be used concerning agricultural practice. Two algorithms are supervised classification methods - Maximum Likelihood (ML) and Supported Vector Machine (SVM). The Pix4Dfields (Magic Tool) classification algorithm and the thresholding method are other methods used. The Kappa coefficient and the overall accuracy determined the accuracy of the individual algorithms. The highest accuracy was achieved by the thresholding method, and the lowest by the Pix4Dfields algorithm. Among the supervised classification methods, SVM achieved higher accuracy than the ML algorithm. In terms of using the methods in practice, the thresholding method proved more effective than supervised classification methods.

---

1. The study was supported by the Internal Grant Agency of the Faculty of AgriSciences at Mendel University in Brno as the research project IGA24-AF-IP-043.

---

2. Thematic Sessions: Short Papers