Low Input Deep Vision Spot Spraying in Iran
Autonomous Spot Spraying with Deep Computer Vision and Microdosing in Low-Input Farms
Conventional spraying in many farms is built on a simple assumption: the entire field surface, even areas without weeds, pests, or disease patches, receives a share of the chemical solution. This logic is operationally simple, but for low-input farms, where input costs, water, labor, and soil health are all under pressure at the same time, it is gradually becoming both an economic and environmental problem. When weeds or disease patches appear in scattered, localized patterns, uniform spraying spends part of the farm’s capital on areas that do not need treatment. Autonomous spot spraying is the technological response to this hidden waste, shifting the logic of pesticide use from field area to the actual presence of a target.
The importance of this shift is not limited to reducing pesticide use. Every liter of solution saved also means a lower carrying load, fewer tank refills, less unnecessary contact between soil or crop and chemical material, and a reduction in part of the operating cost. In existing commercial and research examples, leading technologies use RGB or 3D sensors, neural networks, edge processing, independent nozzle control, and as-applied maps to decide which point should be sprayed and which point should remain untouched. This shift is attractive for low-input farms because it increases input-use efficiency without requiring higher consumption levels. In this model, the machine is no longer merely a tank carrier; it becomes a unit for detection, decision-making, and localized dose application.
Policy pressure has also helped move the sector in this direction. Within its non-binding targets, the European Union has proposed a 50% reduction in the use and risk of chemical pesticides and a 50% reduction in the use of more hazardous pesticides by 2030. Directive 2009/128/EC also highlights the reduction of risks to human health and the environment as a regulatory rationale for the sustainable use of pesticides and places integrated pest management at the center of the discussion. Spot spraying should not be treated as a complete substitute for integrated pest management, but it can serve as one of its technical tools: a tool that brings chemical use closer to the real location, timing, and intensity of need.
From Uniform Spraying to Field-Scale Microdosing
The operational definition of autonomous spot spraying with deep computer vision is a system that receives images or sensor data while moving through the field, detects the target as a weed, diseased plant, or treatable patch, and activates only the nozzle or actuator corresponding to that specific point. In commercial literature, terms such as selective spraying, plant-by-plant spraying, and ultra-high-precision spraying are more common than microdosing. In this discussion, however, microdosing refers to highly localized dose delivery at the centimeter scale or within a small treatment zone. Ecorobotix’s ARA has operationalized this idea with 156 nozzles, 4-centimeter nozzle spacing, and a 6-by-6-centimeter spray pattern. This level of resolution means that the spraying decision is no longer limited to the whole boom or the entire row; it is broken down into smaller controllable units.
At the field scale, this shift means moving from surface management to target management. If the system sprays only the identified plant or patch, the amount of solution used becomes a function of actual target density, not merely a function of field area. A Smart Agricultural Technology study on potatoes showed that variable-rate application, compared with fixed-rate application, reduced spray solution volume by an average of 47% in the weed experiment and 51% in the diseased-plant experiment. The same study reported that cloudy, partly cloudy, and sunny light conditions had no significant effect on spray volume in the weed and diseased-plant experiments, with p-values of 0.93 and 0.75, respectively.
Even so, the more mature and commercial evidence still relates mainly to weeds and herbicides. The technical reason for this focus is clear: weeds usually have more detectable shape, color, location, and visual contrast than many early signs of disease or pest presence, and the link between the detection decision and herbicide activation is also more direct. Evidence related to disease and pests exists in the research record, but compared with weeds, it is more often seen at the level of field trials, prototypes, drones, or controlled environments. Therefore, the short-term commercial use of this technology in open fields relies primarily on weed control and reduced herbicide use.
Technical Architecture of Detection and Spraying
The technical architecture of a successful system consists of four interconnected layers: sensor, detection model, actuator control, and data logging. RGB or 3D sensors capture field images during movement, the neural network detects the target in near real time, the nozzle controller turns the decision into localized opening and closing, and the as-applied map shows which part of the field was actually treated. In its official specifications, ARA reports a maximum speed of 7.2 kilometers per hour and a capacity of up to 4 hectares per hour. These two figures only become meaningful when read alongside target density, tank refill time, working width, calibration accuracy, and actual field conditions.
The main metric in the machine vision layer is not merely raw accuracy; in practice, the ratio of missed targets and unnecessary spraying errors is also decisive. In a university field trial on tobacco, a YOLOv5n-based system achieved an F1 score of 87.2% and a rate of 67 frames per second. The same study reported that selective and variable-rate methods reduced chemical use by up to 60% compared with broadcast spraying. The importance of this result lies in the combination of two criteria: the model must be fast enough to remain synchronized with machine movement, and it must be accurate enough to prevent spot spraying from becoming either incomplete control or unnecessary chemical use.
The strawberry study also clarifies another side of the issue. In that study, VGG-16 performed better than AlexNet and GoogleNet in weed detection, and the completely sprayed weeds target value was reported at 93%. This metric shows what share of the identified targets actually received full spraying and therefore makes the gap between digital detection and mechanical execution visible. The same study emphasized that when the sprayer’s travel speed exceeded 3 kilometers per hour, system performance declined. This finding is important for small and uneven farms, but it should not be generalized to large commercial machines without testing.
Safety standards are also part of the technical architecture, not a separate issue to be added at the end of the design process. ISO 16119-2 defines design and performance requirements for horizontal boom sprayers with the goal of reducing the potential risk of environmental contamination. From this perspective, nozzle accuracy, residual volume, and safe boom design have direct importance. ISO 18497:2018 also covers design principles for highly automated aspects of agricultural machinery and safety information regarding residual risks. When a system is intended to decide, automatically or semi-automatically, where a chemical should be sprayed, mechanical safety, operational data safety, and the ability to stop or correct errors must all be considered simultaneously in the design.
Global Case Studies and Market Formation
The global spot-spraying market has not grown along a single path. Some players offer a complete machine or an ultra-precise boom, while others have pursued a retrofit path for existing sprayers. John Deere’s See & Spray is a leading example of a large commercial machine, and in its 2024 report, the company stated that customers using this technology achieved an average herbicide savings of 59%. An Iowa State study also recorded savings across 415 acres, equivalent to about 168 hectares, including 4,700 gallons of tank solution, or roughly 17,791 liters, and $6,500 in herbicide costs. These figures show that reduced use at the real-farm level can be translated into financial terms, although the amount of savings still depends on weed density, season, and cropping pattern.
Ecorobotix represents another path, because ARA is introduced as an ultra-high-precision sprayer using RGB and 3D vision, plant-by-plant algorithms, and very closely spaced nozzles. In 2025, the company announced that 1,000 ARA units had been sold worldwide and that these machines were being used in Europe, the United States, Canada, and Australia. The company’s official claim of reducing the use of crop protection products and fertilizers by up to 95% depends on weed or target density and should not be read as a fixed average for all farms. The same company announced a total of $150 million in funding in 2025, including a $45 million Series C in 2024 and a $105 million Series D in 2025.
– Dominique Mégret, CEO of Ecorobotix: “Farmers today are facing rising costs and labor shortages.”
One Smart Spray, the joint venture between Bosch and BASF, is an example of an industrial and data-driven alliance. The system entered commercial deployment in Latin America in collaboration with Stara and was introduced in Brazil on the Imperador 4000 Eco Spray sprayer. Its architecture uses Bosch cameras and sensors, artificial intelligence, xarvio Digital Farming, weed distribution maps, as-applied maps, herbicide program recommendations, and automatic documentation. Capabilities such as green-on-brown, green-on-green, and round-the-clock operation with LED lighting show that market competition is not only about nozzle precision, but also about turning spraying into a data-driven and documentable workflow.
On the other side of the market, Greeneye Technology highlights the value of retrofit solutions. The company introduces its artificial intelligence system for real-time weed detection and spraying as a technology that can be installed on existing sprayers, meaning that farmers or service contractors are not necessarily forced to replace the entire machine from the ground up. Greeneye’s partnership with Farmers Business Network provides access to a network of more than 33,000 members and more than 81 million acres, or about 32.8 million hectares, for testing and scaling. Greeneye’s $22 million Series A funding also shows that corporate venture capital sees this field as part of the future of smart mechanization.
The Economics of Replacing Conventional Spraying
The economic model of spot spraying begins with a simple calculation, but it does not end there. The farmer first looks at reduced herbicide or solution use, but return on investment depends on the machine price, software license cost, annual utilization rate, crop value, weed density, operating speed, and access to service. In the John Deere example, in addition to the machine itself, the Unlimited option for See & Spray Premium has been announced at $28,000 per license. This figure shows that the economics of the technology are not limited to buying iron and a boom; software models, feature activation, and data services also form part of the cost of ownership.
The Iowa State data for See & Spray show herbicide savings of $6,500 across 415 acres, equivalent to about $15.7 per acre or $38.8 per hectare. This type of figure is important for economic analysis because it converts savings from a broad percentage into a farm-level calculable unit. However, the savings percentage can change in a field with high weed density, a different cropping pattern, or a narrow application window. Therefore, external figures should be used as inputs for designing a financial model, not as a guaranteed profit promise for every farm.
The industry’s revenue model also has several branches. One path is the sale of a specialized machine or boom along with software, licensing, data services, and maintenance; this path is seen in technologies such as See & Spray and ARA. The second path is retrofitting existing sprayers, which Greeneye has emphasized and which is especially important for low-input farms or spraying service contractors. Retrofit can lower the initial investment barrier, but it still requires mechanical compatibility, precise calibration, operator training, and parts support.
Investments made in this field show that the market is still in the stage of technological formation and scaling, not yet a fully commoditized and mature market. Ecorobotix’s $150 million in funding and Greeneye’s $22 million Series A both point to the fact that developing vision models, manufacturing precision hardware, building sales networks, delivering field services, and conducting field validation are all capital-intensive. For low-input farms, the best model may not be individual ownership of an expensive machine, but rather contract services, cooperative equipment ownership, or phased retrofit. Such a model shifts the entry cost from the individual farm level to the specialized service level and increases the possibility of shared use of the technology.
A Practical Path for Iran and Low-Input Farms
For Iran, a cautious and defensible path does not begin with the claim that the market is ready; it begins with the design of a precise pilot. Global evidence shows that reducing solution use, reducing chemical use, and achieving herbicide savings are possible, but these figures should not be directly converted into investment decisions without testing Iran’s crops, climate, soil, weed density, operator skill, and machine ownership structure. The logical starting point is row crops, vegetables, or crops in which weed-control costs and environmental sensitivity are higher. Such a pilot should be defined from the beginning with metrics for solution consumption, detection accuracy, control quality, operating time, and service cost.
In low-input farms, the value of the technology increases when it is used to replace indiscriminate input consumption, not to make operations more complex without a clear return. If a system can treat only weed patches or diseased plants, less tank water and chemical material are used and cost pressure decreases. The potato example, with 47% and 51% reductions in solution volume, and the tobacco example, with up to 60% reduction in chemical use, can be used to design metrics for Iranian trials. These metrics must be evaluated alongside control quality, final crop yield, and actual operating speed, because reduced use without effective control does not create sustainable economic value for the farmer.
Localization should also be divided into technical components rather than treated as a general slogan. Parts of the system, such as the chassis, boom, tank, pump, decision-making software, and user interface, can be placed on a domestic development path. However, nozzle precision, fast solenoid valve response, industrial cameras, edge processing, accurate positioning, and calibration remain decisive bottlenecks. The ARA experience shows that 4-centimeter nozzle spacing and a 6-by-6-centimeter spray pattern are not merely promotional numbers; they indicate a level of integration among mechanics, vision, and control. Therefore, before making any commercial claim, an Iranian pilot must evaluate spray repeatability, calibration durability, and model performance in real field conditions.
From a policy perspective, this technology in Iran should also be viewed in relation to integrated pest management, reduced unnecessary exposure to pesticides, soil and water protection, and machinery safety standards. The FAO and WHO International Code of Conduct on Pesticide Management highlights health and environmental aspects, while ISO 16119-2 and ISO 18497 are important for the safe design of sprayers and highly automated agricultural machinery, respectively. For a technology holding active in the food security value chain, investment appeal in this field increases when the pilot demonstrates both input reduction and a clear model for service delivery, maintenance, calibration, and operator training. Spot spraying in low-input farms becomes an economic solution only when it passes technical testing and turns into a repeatable field service.
Conclusion
Autonomous spot spraying with deep computer vision is not a simple upgrade to the conventional sprayer; it is a change in the logic of input use on the farm. In this logic, the field surface gives way to the detectable target, and the nozzle is activated only when image data and the detection model confirm the presence of a weed or treatable patch. Research evidence from potatoes, tobacco, and strawberries shows that reduced solution volume, reduced chemical use, and acceptable execution of targeted spraying are possible. Commercial evidence from John Deere, Ecorobotix, One Smart Spray, and Greeneye also shows that the global market has moved beyond pure experimentation and has entered a combination of machinery, software, data, and field services.
The decision-making path for low-input farms must be accompanied by numerical caution and technical precision. Savings percentages, multimillion-dollar investments, and centimeter-level nozzle specifications show the capacity of the technology, but for each country and each farm, they must be translated into ownership models, services, calibration, weed density, and crop value. For Iran, the professional starting point is a crop-focused pilot with clear metrics: a pilot that simultaneously measures solution consumption, input cost, detection accuracy, operating speed, control quality, and safety requirements. Such a path can turn spot spraying from an attractive technology into a reliable investment and environmental decision in knowledge-based agriculture.
