Development of an automated detector and counter for bagworm census

Bagworms (Thyridopteryx ephemeraeformis) are one of the main species of vicious leaf eating insect that is a threat to the oil palm plantations in Malaysia. The economic impact from a moderate bagworm attack of 10%-50% leaf damage may cause 43% yield loss. The population of bagworms if not controlled often increases to above its threshold limits, thereby causing serious losses. Due to this, monitoring and detection of bagworm population in oil palm plantations is required as preliminary steps to ensure proper planning of control actions. A precise bagworm monitoring system is required to overcome recurrence of an outbreak. This study, investigates and explores a thermal imaging technique to detect the bagworms and identifying the bagworms through spectral reflectance properties (bagworm characterization) at different stages of the bagworms life cycle. Furthermore, this study develops an automated bagworm detection and counting technique for bagworm census through image processing analysis and this automated solution is found to be more efficient method in determining the bagworm population when compared to manual census techniques. As for detection, the reflector method was applied to find the reflected apparent temperature and emissivity of the bagworms using thermographic measurement techniques. Then, the experiment on identification of bagworm under thermal imaging is conducted using a thermal infrared camera, T 440 at different sites. It was revealed that the bagworms’ surfaces exhibited emissivity values was recorded approximately at 0.88±0.01 and 0.89±0.02. The statistical results from three rounds of experiments showed that the object/bagworm temperature during the evening, night, and morning were significantly different, p<0.05, as compared to the surrounding/frond temperature, with consideration of emissivity, solar radiation, and snapshot distance. The living and dead bagworm spectral reflectance properties were determined using spectroradiometer, GER1500 under the Visible/Near Infrared and Short-wave Infrared wavelength regions, 350 – 1050 nm, and the results were statistically confirmed using Student’s t- Test with two tailed distributions, principal component analysis and Boxplot Quantiles. The development of an image processing algorithm for detection and counting of Metisa plana Walker, a species of Malaysia’s local bagworm using image segmentation was proposed as it was found to be better than the thermal approach after some preliminary field tests. Color and shape features from the segmented images, combined with deep learning and Faster Region-based Convolutional Neural Networks for real time object detection showed an average detection accuracy, of 40% and 34%, at 30 cm and 50 cm camera distance, respectively. By applying deep convolutional neural network, the percentage of detection increased up to 100% at a camera distance of 30 cm in close condition. The proposed solution was also designed to distinguish between living and dead bagworms using motion detection which results in approximately 73-100% accuracy at a camera distance of 30 cm in the close condition. The fabrication of the prototype was accomplished and field tested. The classification of the larval and pupal stages was carried out by grouping the larval and pupal stages based on their real size; Group 1: larvae stage 1-3, Group 2: larvae stage 4-7 and Group 3: pupal stage. The results showed that the average percentage of the detection accuracy was 87.5% and 78.7%, respectively for the living and dead Group 1 larvae. Meanwhile, the average percentage of the detection accuracy for the living and dead Group 2 larvae was same 79.2%, respectively. As for pupa in Group 3, the result showed that the average percentage of detection accuracy of the prototype to detect the living and dead pupae against manual census was 77% and 75%, respectively. The limitations of this study were determined, such as the camera distance and snapshot condition during image capture were limited at 30 cm and 50 cm, and set in three conditions; open, half open and close condition, damage, brownish leaflet and hole were found as natural limitations, characteristic of the bagworm in term of colour and material of its bag attributed to difficulties to extract the bagworm from its surrounding and SOP for bagworm census. There are several recommendations from this study that have been suggested including the use of hyperspectral imaging to detect bagworms, application of radio frequency to detect live bagworms, open system detection of the bagworms, application of pseudo colour concept and method to detect early stage of bagworm attack.

Text FK 2021 16 - IR.pdf Download (1MB)

Actions (login required)

View Item