2. Opportunities and Possibilities of Developing an Advanced Precision Spraying System for Tree Fruits

Although various machine vision approaches have proved to be very useful and successful in weed detection, crop stress monitoring and yield prediction, etc., challenges still exist, especially in outdoor field conditions. Among the different types of sensors used in machine vision applications, camera sensors are more susceptible to outdoor lighting and weather conditions. The camera vision approaches might be best suited for spot spraying (spray based on site-specific disease or pest pressure) rather variable-rate spraying in the tree fruits. Recent developments of multispectral and hyperspectral cameras offer narrower spectral bands that can penetrate into the canopies and accurately measure the conditions (diseases or pest pressures and stresses because of nutrients) of trees. However, longer data processing time (particularly for hyperspectral cameras) is a major limitation. Current inventions of deep learning-based artificial intelligence systems resulted in good canopy detections and segmentations. Semantic and instance segmentation based deep learning models, including SegNet, U-Net, DeepLab v3, mask-region-based convolutional neural networks (RCNN), and so on, could be the best fit for canopy area segmentation; however, more efforts and extensive investigations are expected to develop an effective and robust system for precision spraying. Apart from camera sensors, ultrasound and laser scanner sensors can overcome light illumination problems usage sound waves and electromagnetic signals, respectively, as inputs to measure detection distance and gather canopy structure parameters. As compared to ultrasonic and camera sensors, laser scanner sensors are more accurate and reliable in collecting crop structure information, are not affected by weather conditions, and can guide the sprayer unit efficiently and effectively to apply pesticides on target.

The application of laser scanners has risen during the last decade in multiple agriculture applications, from automatic tree height measurement to site-specific pesticide application. However, challenges are still presented, especially in site-specific or variable-rate spraying in tree fruit orchards. Accurate canopy information measurements require automatic adjustment of laser sensor orientation during data acquisition, especially in conditions of different sloping and uneven terrains. Therefore, a real-time terrain monitoring system may need to be incorporated along with a sensing system for precision spraying. Terrain monitoring systems can be developed using a positioning sensor and a real-time kinematic global positioning system (RTK-GPS). An inertial measurement unit (IMU) sensor can be incorporated into the spraying system to monitor the terrain conditions in real-time and adjust the laser orientation accordingly. Studies reported IMU sensors could minimize positioning errors of the laser from terrain variabilities ^[20][21]. The RTK-GPS will locate the uneven terrain position and guide the spraying system to follow the co-ordinate where the necessary adjustment is needed for accurate canopy information. Few studies have considered uneven terrain information for correcting tree canopy information ^[22][23], so more studies need to be conducted to implement real-time correction of tree canopies for accurate spray deposition.

Laser scanner sensors only help with measuring canopy structure information, but a successful spraying operation requires a series of tasks to be completed accurately, including maintaining constant travel speed, adjustments of the fan speed at the sprayer unit, assessment of real-time weather conditions, prediction of droplet sizes and some other factors. Auto-guidance systems that automatically steer the sprayer throughout the orchard blocks could help maintain constant travel speeds. Crossflow fan sprayers reduce off-target spray deposition because they have more appropriate nozzle positions and uniform air delivery. An automatic damper system can be added at the sprayer fan’s air inlet (back of the sprayer) to control the airflow based on tree canopy information (Figure 7Figure 2). A mathematical model will need to be developed to calculate the airflow required for sending the spray droplet to the desired canopies.

Position of damper installation for precision sprayer airflow control.

Real-time weather assessment is also important to reduce off-target deposition. Wind can easily blow spray from targeted to non-targeted regions. The real-time control of nozzles based on weather data is not currently practical during spray operation. The future sprayer also needs to include a droplet size prediction model to determine the diameter of droplets based on the characteristics of the tree canopies during spray operations. Electrostatic sprayers have provided improved deposition of specific droplet sizes and may be a good aid to develop next-generation sprayer systems ^[24]. Electrostatic sprayers apply a positive charge to liquids as they pass through the nozzle. The positively charged liquids are attracted to negatively charged surfaces, which allows for efficient coating of hard nonporous surfaces. A 44% increase in mean deposition using the electrostatic system reported by Pascuzzi and Cerruto ^[25]. Some other factors including canopy conditions and spray droplet retention depend heavily on the tree canopies and cannot always be addressed by the spraying unit. However, the nozzle effect can be resolved by matching appropriate nozzles with electronically controlled valves.

The future of precision spraying systems will focus on integrating a suite of robotic and sensor-guided spraying systems with fruit tree growth and structure information to ensure accurate dose of pesticide delivery and reduce off-target deposition. A wide variety of sensor-based spraying technologies are available to measure the amount of spray volume needed; however, it is also important to assess the durability of these technologies so that a variety of sensors can be added at reasonable price for the farmers. Concurrently with sensor development, site-specific disease, pests, and stress monitoring using technologies such as deep learning and remote sensing will improve the functioning of current sensor-based spraying systems. Fusing/integrating different sensors could help to add more decision-making features for more precise pesticide applications in tree fruits. Future spraying systems must also handle diverse agrochemical formulations beyond that of the standard aqueous mixtures currently used. Although tractor-based spraying systems are mostly used for agrochemical applications, an alternative UAV based application method could displace traditional sprayers. UAVs have been tested in the field crops for disease scouting, stress monitoring, and spraying, and its preliminary results for spraying is showing promise. But most of the current UAV based spraying technologies may not be suitable for spraying in large scale tree fruit orchards due to their limited payloads and variable spray deposition capability. More research and development into UAV swarms and large capability UAV similar to unmanned helicopters could result in more effective system for commercial/large scale fruit orchards. Detection of disease and stress, pest pressure are other avenues for their use in tree fruits production besides variable-rate spraying. Therefore, the integration of multiple sensing and control systems will be the key to using advanced precision sprayers in the future.