Application of RPA and AI in the Identification of Pests in Cotton Plantations in the Brazilian Cerrado

Amaranthus palmeri is a weed in North America that significantly reduces productivity, grows rapidly (25–50mm/day), and produces approximately 500,000 seeds/plant. In Brazil, it was detected in Mato Grosso in 2015. It is now found in other states, threatening soybean and cotton crops. Conventional inspection cannot cover large areas. To address this challenge, the research focused on using RPA and AI to develop an "open source" platform to map, identify, and monitor crops. To obtain the data, a Matrice 350 – RTK, a DJI P1 aerial photogrammetry camera, and a GNSS DA 2 were used. Two data collection stages were carried out to monitor the phenological process. The images were pre-processed with the software "Metashape." The samples were used to train a neural network, and processing was carried out with the GeoLearning plugin; finally, testing and validation were conducted on areas not used for training to avoid bias in the results. For the final flight adjustments, several tests were conducted to obtain the best image, given weed size and possible mixing with other species. Among the Deep Learning architectures, YOLO 11 was used. GeoLearning was developed in three main modules: sample collection, training, application, and detection. The plugin was built in PyQt 5 and integrated into QGIS using GDAL, Ultralytics, and OpenCV. After developing and testing GeoLearning and collecting samples, the researchers trained the model on Google Colab. The performance metrics (precision—95%, recall—93%, and MAP @ 0.5—99%) showed that the model can reliably identify the target species and generate a final report that includes the identified elements with their respective coordinates.