In this paper, we present a multi-modal dataset for obstacle detection in agriculture. The dataset comprises approximately 2 h of raw sensor data from a tractor-mounted sensor system in a grass mowing scenario in Denmark, October 2016. Sensing modalities include stereo camera, thermal camera, web camera, 360 ${}^{\circ }$ camera, LiDAR and radar, while precise localization is available from fused IMU and GNSS. Both static and moving obstacles are present, including humans, mannequin dolls, rocks, barrels, buildings, vehicles and vegetation. All obstacles have ground truth object labels and geographic coordinates.

Optimal fertilization of clover-grass fields relies on knowledge of the clover and grass fractions. This study shows how knowledge can be obtained by analyzing images collected in fields automatically. A fully convolutional neural network was trained to create a pixel-wise classification of clover, grass, and weeds in red, green, and blue (RGB) images of clover-grass mixtures. The estimated clover fractions of the dry matter from the images were found to be highly correlated with the real clover fractions of the dry matter, making this a cheap and non-destructive way of monitoring clover-grass fields. The network was trained solely on simulated top-down images of clover-grass fields. This enables the network to distinguish clover, grass, and weed pixels in real images. The use of simulated images for training reduces the manual labor to a few hours, as compared to more than 3000 h when all the real images are annotated for training. The network was tested on images with varied clover/grass ratios and achieved an overall pixel classification accuracy of 83.4%, while estimating the dry matter clover fraction with a standard deviation of 7.8%.

Weeding operations represent an effective approach to increase crop yields. Reliable and precise weed detection is a prerequisite for achieving high-precision weed monitoring and control in precision agriculture. To develop an effective approach for detecting weeds within the red, green, and blue (RGB) images, two state-of-the-art object detection models, EfficientDet (coefficient 3) and YOLOv5m, were trained on more than 26,000 in situ labeled images with monocot/dicot classes recorded from more than 200 different fields in Denmark. The dataset was collected using a high velocity camera (HVCAM) equipped with a xenon ring flash that overrules the sunlight and minimize shadows, which enables the camera to record images with a horizontal velocity of over 50 km h-1. Software-wise, a novel image processing algorithm was developed and utilized to generate synthetic images for testing the model performance on some difficult occluded images with weeds that were properly generated using the proposed algorithm. Both deep-learning networks were trained on in-situ images and then evaluated on both synthetic and new unseen in-situ images to assess their performances. The obtained average precision (AP) of both EfficientDet and YOLOv5 models on 6625 synthetic images were 64.27% and 63.23%, respectively, for the monocot class and 45.96% and 37.11% for the dicot class. These results confirmed that both deep-learning networks could detect weeds with high performance. However, it is essential to verify both the model’s robustness on in-situ images in which there is heavy occlusion with a complicated background. Therefore, 1149 in-field images were recorded in 5 different fields in Denmark and then utilized to evaluate both proposed model’s robustness. In the next step, by running both models on 1149 in-situ images, the AP of monocot/dicot for EfficientDet and YOLOv5 models obtained 27.43%/42.91% and 30.70%/51.50%, respectively. Furthermore, this paper provides information regarding challenges of monocot/dicot weed detection by releasing 1149 in situ test images with their corresponding labels (RoboWeedMap) publicly to facilitate the research in the weed detection domain within the precision agriculture field.

Soil surface measurements play an important role in the performance assessment of tillage operations and are relevant in both academic and industrial settings. Manual soil surface measurements are time-consuming and laborious, which often limits the amount of data collected. An experiment was conducted to compare two approaches for measuring and analysing the cross-sectional area and geometry of a furrow after a trailing shoe sweep. The compared approaches in this study were a manual pinboard and a Light Detection and Ranging (LiDAR) sensor. The experiments were conducted in coarse sand and loamy sand soil bins exposed to three levels of irrigation. Using the LiDAR, a system for generating 3D scans of the soil surface was obtained and a mean furrow geometry was introduced to study the geometrical variations along the furrows. A comparison of the cross-sectional area measurements by the pinboard and the LiDAR showed up to 41% difference between the two methods. The relation between irrigation and the resulting furrow area of a trailing shoe sweep was investigated using the LiDAR measurements. The furrow cross-sectional area increased by 11% and 34% under 20 mm and 40 mm irrigation compared to non-irrigated in the coarse sand experiment. In the loamy sand, the cross-sectional area increased by 17% and 15% by irrigation of 20 mm and 40 mm compared to non-irrigated measured using the LiDAR.

Grain price differences due to protein content can have economic effects on the farm as well as environmental effects when alternative protein sources are imported. Grain protein variability can vary from year to year due to environmental factors and can be addressed by site-specific management practices. Alternatively, it can be addressed at harvest time by selective harvest. Agricultural autonomous robots can accurately follow alternative harvesting routes that are subject to grain quality maps, making them suitable choices for selective harvest. This study addresses therefore the potential revenue of selective harvest performed by the route planner of an autonomous field robot. The harvest capacity and potential economic revenues of selective harvest in a Danish context were studied for a set of 20 winter wheat fields with four hypothetical scenarios. The results showed significant differences in harvest capacity between conventional and selective harvest. Even though in some scenarios selective harvest did not require notable additional harvest times, the cost–benefit analysis showed small economic returns of up to 46 DKK ha⁻¹ for the best scenarios, and for most cases losses up to 464 DKK ha⁻¹. Additionally, the location of the high protein content areas has great influence on the profitability of selective harvest.