Evaluating The Performance of a Mite Dispensing System for Biological Control of Chilli Thrips in Strawberry Production in Florida

Strawberry pests, specifically chilli thrips (Scirthothrips dorsalis), contribute to significant economic losses for Florida strawberry growers. Newer approaches for pest control consider the use of biological predators like predatory mites (Amblyseius swirskii) as an alternative for chemical pesticides because of the dangers and concerns posed by chemical pesticides. This project demonstrates the potential of using a computer vision and robotics system to assist in autonomously releasing predatory mites on all the strawberry plants in a test field. The system used vermiculite substrate in the place of live predators since the predators are grown in vermiculite substrate. The plant detection model was trained using YOLOv5 and resulted in recall and precision values greater than 97% for training and test datasets. The mean average precision for Intersection over Union between 0.5 and 0.95 was greater than 88 % for both the training and test datasets. The plant detection algorithm was combined with a vermiculite dispensing system and mounted on a remote-controlled ground vehicle to test in a real strawberry field. The vehicle was set at a speed between 0.152 – 0.157 m/s (30 – 31 ft/min) and navigated through 325 strawberry plants. The system had a release accuracy of 69.3 %. The reduced accuracy was primarily due to several factors: an assumption about the arrangement of strawberry plants that did not align with actual conditions, limitations of the ground vehicle hardware, constraints of the computer and microcontroller hardware, and varying environmental conditions. Future experiments will address these limitations as well as add an autonomous capability and a double strawberry row functionality to the system.