The functionality and features of AI automated detector and counter, Oto-BaCTM for bagworm census in oil palm plantation

Abstract

In Malaysia, the bagworm species Metisa plana Walker (Lepidoptera: Psychidae) has been recognized as one of the major insect pests attacking oil palm. Bagworm attacks, if left untreated, have contributed to up to 43% yield loss, and in 2020, the loss of fresh fruit bunches due to these attacks has been chronicled at approximately RM 180 million. Due to this critical situation, it has become compulsory to frequently monitor bagworm infestations in affected areas. Manual census methods, involving the counting of bagworm populations on fronds, have often yielded imprecise data due to human errors, such as overestimating or fabricating data, which have hindered the planning of effective control actions in infested areas. Recognizing the necessity for improved census operations and outcomes, this technology has aimed to create and implement a specialized machine vision system incorporating image processing algorithms tailored to its functions. The device, known as the Automated Bagworm Counter or trademarked as Oto-BaCTM, has stood as the world's inaugural prototype in its category. The software has been configured to operate through Graphic Processing Unit (GPU) computation with the TensorFlow/Teano library, utilizing a trained dataset. Oto-BaCTM has employed a standard camera and self-designed deep learning (DL) algorithms, encompassing motion tracking and false color analysis, to identify both living and deceased larvae and pupae of M. plana. It has further tallied the respective populations of living and deceased larvae and pupae per frond, classifying them into three primary groups or sizes. This automated device has proven to be straightforward, precise, and user-friendly for bagworm detection and counting on palm leaflets. The technology has relied on advanced deep learning with Faster R-CNN methodology for real-time object detection. Although the instrument has not undergone enumeration testing, it has been crafted to augment bagworm counting efficiency in fieldwork, marking it as the pioneer in its domain. Oto-BaCTM's efficacy and detection precision have been authenticated through a series of field trials conducted at two distinct oil palm plantations afflicted by M. plana infestations. Through the integration of infrared sensors and image processing algorithms, this device has been effectively deployed by plantation workers to oversee bagworm populations in their fields, ultimately leading to enhanced yields. This device has exhibited substantial potential for utilization and commercialization, particularly in aiding workers engaged in census-related endeavors.