<p>Current methods of broadcast herbicide application cause a negative environmental and economic impact. Computer vision methods, specifically those related to object detection, have been reported to aid in site-specific weed management procedures to target apply herbicide on per-weed basis within a field. However, a major challenge to developing a weed detection system is the requirement for properly annotated training data to differentiate between weeds and crops under field conditions. This research involved creating an annotated database of weeds by using UAS-acquired imagery from corn and soybean research plots located in North-central Indiana. A total of 27,828 RGB; 108,398 multispectral; and 23,628 thermal images, were acquired using FLIR Duo Pro R sensor that was attached to a DJI Matrice 600 Pro UAS. An annotated database of 306 RGB images, organized into monocot and dicot weed classes, was used for network training. Two Deep Learning networks namely, DetectNet and You Only Look Once version 3 (YOLO ver3) were subjected to five training stages using four annotated image sets. The precision for weed detection ranged between 3.63-65.37% for monocot and 4.22-45.13% for dicot weed detection. This research has demonstrated a need for creating a large annotated weed database for improving precision of deep learning algorithms through better training of the network.</p>

Effective weed control is crucial, especially for herbicide-resistant species. Machine vision technology, through weed detection and localization, can facilitate precise, species-specific treatments. Despite the challenges posed by unstructured field conditions and weed variability, deep learning (DL) algorithms show promise. This study evaluated thirteen DL-based weed detection models, including YOLOv5, RetinaNet, EfficientDet, Fast RCNN, and Faster RCNN, using pre-trained object detectors. RetinaNet (R101-FPN) achieved the highest accuracy with a mean average precision (mAP@0.50) of 79.98%, though it had longer inference times. YOLOv5n, with the fastest inference (17 ms on Google Colab) and only 1.8 million parameters, achieved a comparable 76.58% mAP@0.50, making it suitable for real-time use in resource-limited devices. A prototype using YOLOv5 was tested on two datasets, showing good real-time accuracy on In-season data and comparable results on Cross-season data, despite some accuracy challenges due to dataset distribution shifts.

Le contexte agricole actuel vise à réduire l’utilisation des produits phytosanitaires sur les parcelles. Dans ce cadre, la gestion des adventices consommant de grandes quantités d’herbicides est devenue une problématique majeure. Afin de mettre en place un outil de gestion localisée des adventices par drone, cette thèse étudie l’adaptation du système d’acquisition (drone + dispositif multispectral) actuellement proposé par AIRINOV à la détection des adventices sur des cultures sarclées. La chaîne d’acquisition a été modélisée afin d’évaluer l’impact de différents paramètres du modèle (filtres optiques et résolution spatiale) sur la qualité de la détection des adventices. Des orthophotographies et images ortho-rectifiées ont été acquises à l’aide d’un capteur multispectral (4 et 8 filtres) à des résolutions spatiales de 6 mm et 6 cm. Plusieurs méthodes de localisation des adventices adaptées à l’étude de ces images ont été développées. Elles reposent sur 1) l’analyse de la distribution spatiale de la végétation (détection de rang par la transformée de Hough et analyse de forme), 2) la classification spectrale des pixels (méthodes supervisées : LDA, QDA, distance de Mahalanobis, SVM). Enfin, une classification spectrale basée sur un apprentissage issu des informations spatiales été proposée, améliorant ainsi la détection des adventices.Des cartes d’infestation des parcelles et de préconisation en pulvérisation localisée ont alors été créées. / The agricultural framework aims to reduce pesticide use on fields. Weed management, which is highly herbicide consuming, became a great issue. In order to develop a weed management service using UAV, this PhD dissertation studies how to adapt the acquisition system (UAV + multispectral camera) developed by AIRINOV to detect weeds in row crops. The acquisition chain was modeled to assess some of its parameters (optical filters and spatial resolution) impact on weed detection quality. Orthoimages and orthorectified images were created using a multispectral camera (4 to 8 filters) with 6 mm to 6 cm spatial resolutions. Several weed location methods were specifically developed to study multispectral images acquired by UAV. They are based on 1) the analysis of vegetation spatial distribution (row detection using the Hough transform and shape analysis), 2) spectral classification of pixels (supervised methods: LDA, QDA, Mahalanobis distance, SVM). In order to improve weed detection, a spectral classification based on training data deduced from spatial analysis was then proposed.Weed infestation maps and recommendation for spot spraying applications were then produced.

Modern weeding is predominantly carried out by spraying whole fields with toxic pesticides, a process that accomplishes the main goal of eliminating weeds, but at a cost of the local environment. Weed management systems based on AI solutions enable more targeted actions, such as site-specific spraying, which is essential in reducing the need for chemicals. To introduce sustainable weeding in Swedish farmlands, we propose implementing a state-of-the-art Deep Learning (DL) algorithm capable of instance segmentation for remote sensing of weeds, before coupling an automated sprayer vehicle. Cereals have been chosen as the target crop in this study as they are among the most commonly cultivated plants in Northern Europe. We used Unmanned Aerial Vehicles (UAV) to capture images from several fields and trained a Mask R-CNN computer vision framework to accurately recognize and localize unique instances of weeds among plants. Moreover, we evaluated three different backbones (ResNet-50, ResNet101, ResNeXt-101) pre-trained on the MS COCO dataset and through transfer learning tuned the model towards our classification task. Some well-reported limitations in building an accurate model include occlusion among instances as well as the high similarity between weeds and crops. Our system handles these challenges fairly well. We achieved a precision of 0.82, recall of 0.61, and F1 score of 0.70. Still, improvements can be made in data preparation and pre-processing to further improve the recall rate. All and all, the main outcome of this study is the system pipeline which, together with post-processing using geographical field coordinates, could serve as a detector for half of the weeds in an end-to-end weed removal system. / Site-specific Weed Control in Swedish Agriculture