Abstract
The ability of thrips to infect plants with viruses, oviposition, and feeding makes it one of the most damaging agricultural pests in the world. Due to their invasiveness, many pest species pose serious challenges to growers of a wide variety of crops especially Mango in both open fields and greenhouses globally. For detecting various pests, sticky paper traps have been widely used. It is essential to inspect sticky paper traps as part of an integrated pest management (IPM) process. Identification and counting the pests caught in the traps is a laborious process. A timely and effective strategy is therefore required to resolve the issue. This study proposes and develops a complete and feasible integrated pest detection and monitoring system (IPDMS) by combining ultra-high resolution pest monitoring device coupled with deep learning based object detection model. The developed system was able to collect sticky paper images with a resolution of 64 megapixels and AWS database for analysis. Test results from two Mango greenhouses shows that the model can detect thrips with 90% accuracy and an F1-score of 0.95. Additionally, a web-based user interface was developed for providing visual image recognition results and thrip population analysis. Thus, the proposed IPDMS addresses the significant problems with pest detection systems by providing real-time information on pest occurrences in greenhouses. Thus, it makes pest detection and monitoring in agriculture efficient.
Funding Information
  • National Science and Technology Council (MOST 111-2221-E-007-063-MY3, MOST 111-2622-8-007-008-TE2)

This publication has 24 references indexed in Scilit: