<![CDATA[This study investigates the development of a detection and repellent system for Lonchura birds in paddy fields by assigning object detection inference to the sensor level (in-sensor computing) using a Sony IMX500 intelligent camera module combined with a Raspberry Pi 5 and a motorized pan-tilt mechanism. The applied methodologies include picture dataset compilation and augmentation, transfer learning utilizing an object identification architecture, actuator control implementation, and quantitative evaluation of model performance alongside specimen behavioral response testing. The results demonstrate a precision improvement to 86.4% and a mean Average Precision (mAP) of 87.6% at 0.50; the inference rate is recorded at 25 to 30 frames per second; automatic detection is effective up to 200 cm, along with a response to auditory stimuli for avoidance. The conclusion says that the integrated architecture significantly reduces false detections and delays, hence improving the effectiveness of non-lethal insect management.]]>
Copyrights © 2026
