Object detection models, such as those in the YOLO family, are generally effective identifying individual weeds, but their performance can be limited by occlusion of target structures in high density scenes. These models are typically trained on images with low weed densities, where individual plants are clearly visible and easy to annotate, yet they are used in field conditions where areas of high density and dense vegetation may occur. Greenhouse experiments were conducted at the University of Florida, Wimauma, FL, to evaluate how purple nutsedge (Cyperus rotundus L.) density impacts the performance of a YOLOv8 model. A greenhouse density test dataset was developed by collecting 480 images of twelve densities ranging from 7 to 331 plants m-2, at transplant and at 3, 6, 10, and 17 days after transplanting. An independent dataset of 2,221 field and greenhouse images was used for the training of a YOLOv8 extra-large model to detect C. rotundus. A logistic sigmoidal model was fitted to evaluate the F1 score as a function of increasing C. rotundus density at each data collection time point. The F1 score showed sigmoidal relationships with density at all time points, exceeding 0.9 at low densities but dropping sharply to near 0 at the highest density and later time points. Performance decline was primarily driven by increased false negatives as density and occlusion increased, with minimal contribution from false positives except at the highest densities. Density thresholds for optimal performance (F1 ≥ 0.90) decreased from 157 to 86 plants m-2 as canopy coverage increased, while marginal performance (F1 = 0.50) dropped from 322 to 140 plants m-2. Our findings suggest that object detection models for C. rotundus are strongly influenced by increased occlusion and morphological changes resulting from greater plant proximity and canopy coverage in high density scenes.

Precision weed detection and mapping in vegetable crops are beneficial for improving the effectiveness of weed control. This study proposes a novel method for indirect weed detection and mapping using a detection network based on the You-Only-Look-Once-v8 (YOLOv8) architecture. This approach detects weeds by first identifying vegetables and then segmenting weeds from the background using image processing techniques. Subsequently, weed mapping was established and innovative path planning algorithms were implemented to optimize actuator trajectories along the shortest possible path. Experimental results demonstrated significant improvements in both precision and computational efficiency compared with the original YOLOv8 network. The mean average precision at 0.5 (mAP50) increased by 0.2, while the number of parameters, giga floating-point operations per second (GFLOPS), and model size decreased by 0.57 million, 1.8 GFLOPS, and 1.1 MB, respectively, highlighting enhanced accuracy and reduced computational costs. Among the analyzed path planning algorithms, including Christofides, Dijkstra, and dynamic programming (DP), the Dijkstra algorithm was the most efficient, producing the shortest path for guiding the weeding system. This method enhances the robustness and adaptability of weed detection by eliminating the need to detect diverse weed species. By integrating precision weed mapping and efficient path planning, mechanical actuators can target weed-infested areas with optimal precision. This approach offers a scalable solution that can be adapted to other precision weeding applications.

Site-specific weed management (on the scale of a few meters or less) has the potential to greatly reduce pesticide use and its associated environmental and economic costs. A prerequisite for site-specific weed management is the availability of accurate maps of the weed population that can be generated quickly and cheaply. Improvements and cost reductions in unmanned aerial vehicles (UAVs) and camera technology mean these tools are now readily available for agricultural use. We used UAVs to collect aerial images captured in both RGB and multispectral formats of 12 cereal fields (wheat [Triticum aestivum L.] and barley [Hordeum vulgare L.]) across eastern England. These data were used to train machine learning models to generate prediction maps of locations of black-grass (Alopecurus myosuroides Huds.), a prolific weed in UK cereal fields. We tested machine learning and data set resampling methods to obtain the most accurate system for predicting the presence and absence of weeds in new out-of-sample fields. The accuracy of the system in predicting the absence of A. myosuroides is 69% and its presence above 5 g in weight with 77% accuracy in new out-of-sample fields. This system generates prediction maps that can be used by either agricultural machinery or autonomous robotic platforms for precision weed management. Improvements to the accuracy can be made by increasing the number of fields and samples in the data set and the length of time over which data are collected to gather data across the entire growing season.

The presence of a soil seedbank facilitates the persistence of annual weed species in arable fields. Soil weed seedbank is replenished by many sources, but the largest one is the seeds produced by uncontrolled late-season weed escapes. The estimation of weed seed production potential from late-season escapes may allow farmers to make appropriate management decisions to minimize seedbank replenishment. The objective of this research was to evaluate the feasibility of using unmanned aerial vehicle–based RGB and multispectral imagery for estimating seed rain potential in late-season weed escapes in crop fields. Three case studies were used to capture images of weed escapes before crop harvest: common waterhemp [Amaranthus tuberculatus (Moq.) Sauer] in soybean [Glycine max (L.) Merr.], Palmer amaranth [Amaranthus palmeri (S.) Watson] in cotton (Gossypium hirsutum L.), and johnsongrass [Sorghum halepense (L.) Pers.] in soybean. Randomly selected quadrats with different density gradients of weed escapes were sampled at the time of crop maturity. High-resolution RGB and multispectral images of the experimental area were collected using drones immediately before ground sample collection. Normalized difference vegetation index (NDVI), excess green index (ExG), and canopy volume estimates derived from canopy height models were used to obtain weed biological measurements (biomass and seed production). Among the indices investigated, NDVI and ExG had very strong correlations (0.71 to 0.97) with weed biomass. No specific remote sensing variable was ideal across the three cases examined here, suggesting that a generalized remote sensing approach may not offer robust estimations and case-specific applications are imperative. Nonetheless, drone imagery is a powerful tool for estimating seed production from uncontrolled weed escapes and assisting with management decision making.

Glyphosate-resistant (GR) and glyphosate-tolerant weeds cause considerable yield losses and represent a growing threat to soybean production systems. Despite the relevance of this topic, few studies have evaluated the dispersal of these species in Brazil. The objective of this study was to evaluate the dispersal and frequency of known GR and glyphosate-tolerant weeds in soybean-producing microregions. A total of 2,481 interviews were conducted in different regions of Brazil. The interviews were stratified among 20 edaphoclimatic microregions (ECRs) to cover all of the country’s soybean-producing regions. A minimum number of interviews was estimated to generate a margin of error of ≤10% within the ECRs and ≤5% in the country. The values of the farmers’ responses were extrapolated to the total soybean production area of each ECR and the country as a whole, and the absolute values of each response were normalized as percentage values. The dispersal and management data demonstrate a loss of efficiency of glyphosate-resistance technology. Species that are naturally tolerant to glyphosate such as goosegrass, Commelina spp., and Ipomoea spp. had a greater presence in the ECRs, as did the resistant biotypes, particularly Conyza spp. and sourgrass, due to the large area cultivated with GR soybean, where glyphosate has been used with high frequency.

Soil application of herbicides for preemergence (PRE) weed control in grain sorghum is vital to control weeds. Efficacy of soil-applied herbicides is impacted by herbicide adsorption which is influenced by soil organic matter (SOM) and texture. With precision agriculture technologies, variable rate applications (VRA) can be utilized to maximize herbicide effectiveness. In 2016, algorithms were developed for two locations to use VRA of two tank mixed herbicides based on SOM and soil electrical conductivity (EC) collected by a Veris MSP3 system. Drone imagery provided an effective way to evaluate the efficacy of herbicide applications along with visual assessment. VRA applications of herbicide tank mixes provided equal weed control compared to flat rate applications.

In this case study, we evaluated a point-mapping method for simultaneously collecting data while controlling three invasive woody plant species: black locust, Chinese privet, and hardy orange. The study in Arkansas Post National Memorial included seven project areas ranging in size from 2.7 to 27.3 ha and spanned six field seasons (2010 to 2015). The control techniques varied depending on plant size and always included the application of herbicide, which also varied over the course of the study to include glyphosate, imazapyr, and triclopyr. Each person responsible for controlling plants simultaneously collected global positioning system point data to estimate the foliar cover of the plants treated. The resulting data demonstrated evidence of decreases in all three plant species in most project areas during the 6-yr period. Initial increases in area treated for some species–area combinations reflected differences in the preliminary efforts required to control invasive plants in entire project areas, but by 2012 six of seven project areas were treated in their entirety. Despite a high level of reduction, in some cases, the plants persisted at low levels even during the sixth year of the project. Our findings support the ability of this method to granularly detect changes in plant abundance while simultaneously controlling invasive plants. With several acknowledged limitations, this streamlined project-based monitoring approach provides data that allow managers to assess the effectiveness of weed control treatments.

The objective of this research was to explore the influence that weed decision threshold (DT; expressed as plants m−2), weed spatial distribution patterns, and spatial resolution of sampling have on potential reduction in herbicide use under site-specific weed management. As a case study, a small plot located in a typical corn field in central Spain was used, constructing very precise distribution maps of the major weeds present. These initial maps were used to generate herbicide prescription maps for each weed species based on different DTs and sampling resolutions. The simulation of herbicide prescription maps consisted of on/off spraying decisions based on information from two different approaches for weed detection: ground-based vs. aerial sensors. In general, simulations based on ground sensors resulted in higher herbicide savings than those based on aerial sensors. The extent of herbicide reductions derived from patch spraying was directly related to the density and the spatial distribution of each weed species. Herbicide savings were potentially high (up to 66%) with relatively sparse patchy weed species (e.g., johnsongrass) but were only moderate (10 to 20%) with abundant and regularly distributed weed species (e.g., velvetleaf). However, DT has proven to be a key factor, with higher DTs resulting in reductions in herbicide use for all the weed species and all sampling procedures and resolutions. Moreover, increasing DT from 6 to 12 plants m−2 resulted in additional herbicide savings of up to 50% in the simulations for johnsongrass and up to 28% savings in the simulations for common cocklebur. Nonetheless, since DT determines the accuracy of patch spraying, the consequences of using higher DTs could be leaving areas unsprayed, which could adversely affect crop yields and future weed infestations, including herbicide-resistant weeds. Considering that the relationship between DT and accuracy of herbicide application depends on weed spatial pattern, this work has demonstrated the possibility of using higher DT values in weeds with a clear patchy distribution compared with weeds distributed regularly.

Early detection of an invading nonindigenous plant species (NIS) may be critical for efficient and effective management. Adaptive survey sampling methods may provide unbiased sampling for best estimates of distribution of rare and spatially clustered populations of plants in the early stages of invasion. However, there are few examples of these methods being used for nonnative plant surveys in which travelling distances away from an initial or source patch, or away from a road or trail, can be time consuming due to the topography and vegetation. Nor is there guidance as to which of the many adaptive methods would be most appropriate as a basis for invasive plant mapping and subsequent management. Here we used an empirical complete census of four invader species in early to middle stages of invasion in a management area to assess the effectiveness and efficiency of three nonadaptive methods, four adaptive cluster methods, and four adaptive web sampling methods that all originated from transects. The adaptive methods generally sampled more NIS-occupied cells and patches than standard transect approaches. Sampling along roads only was time-efficient and effective, but only for species with restricted distribution along the roads. When populations were more patchy and dispersed over the landscape the adaptive cluster starting at the road generally proved to be the most time-efficient and effective NIS detection method.

Managing weed infestations in a spatially precise manner requires accurate and cost-effective weed identification techniques. The goal of our research was to quantify the accuracy of continuous weed presence–absence maps and assess how management based on those maps may affect producer net returns. Each continuous sampled map covered the entire field and contained vector polygons labeled as either wild oat presence or wild oat absence. The accuracy of the continuous wild oat maps at each sampling time was determined from georeferenced quadrats of wild oat densities. The accuracy of the continuous wild oat seedling maps ranged from 48.3 to 87.1% among the six site-years. The accuracy of the wild oat seedling maps improved by at least 8% when a 10-m buffer was included around areas mapped as wild oat presence. The accuracy of continuous wild oat panicle maps from the combine at harvest ranged from 65.8 to 90.9% among the six site-years. The variation in accuracy for the wild oat seedling maps among sites was greater than the accuracy of the panicle maps. Net returns ($ ha−1) for four site-years were calculated and compared for four possible weed management approaches on each field. A site-specific herbicide application to areas mapped as wild oat presence always generated higher net returns than a herbicide application over the entire field for four sites. A site-specific herbicide application to areas mapped as wild oat presence plus a surrounding 10-m buffer area only resulted in a higher net return in one of the 12 site-years compared with a site-specific herbicide application without the 10-m buffer. This site had the lowest (48.3%) wild oat seedling map accuracy, and uncontrolled wild oat had a high-yield effect. This research indicates that using a continuous weed sampling method based on presence or absence for site-specific herbicide application can be profitable over a herbicide application to the entire field, even with the associated technology cost and seedling map errors.

The distribution and abundance of invasive species can be strongly influenced by habitat suitability and by corridors that facilitate dispersal. We synthesize results from a large-scale invasive plant survey with a patch-scale expansion experiment. The large-scale survey involved transects up to 250 m away from of all roads in a 32,000 ha forest. The patch experiment involved initiating invasions in different habitat types (roadside, wetland, disturbed, and intact forests), and then fitting statistical models to patch spread rates. The large-scale survey highlighted the importance of roads in predicting the presence of invasive plants, also revealing that one invasive plant, Microstegium vimineum, has spread rapidly since its purported introduction in 1994. The patch-scale experiments focused on Microstegium and demonstrated that spread rates are higher in roadsides than in forested and wetland patches, even in the absence of major disturbances. These results highlight the importance of landscape features when designing prevention and management practices aimed at limiting invasive plant abundance and spread.