<![CDATA[There is an emerging requirement for early detection of pests in the field concerning agricultural yield and quality improvement. Traditional methods often result in a loss of the desired outcome due to delayed intervention and increased crop losses. This work focuses on establishing an integrated pest detection system using a hybrid model that combines MobileNet and Faster R-CNN, optimized for real-time performance at the edge. Additionally, LoRa-based data transmission was employed, along with a GIS interface for monitoring. The system is further tested with the diverse dataset of 4,736 images representing common rice pests. It included lightweight feature extraction with precision object detection, as it produced the lowest loss among other models tested. Further implementation is made on a Raspberry Pi, which shows optimal performance in detecting at a distance of 15 cm and with 100 lux of lighting. LoRa communication was adopted for effective data transmission with low power consumption and extensive coverage, while the GIS interface enabled real-time monitoring of pests in space and time. Field tests demonstrated that this system achieved very high accuracy, rapid response, and was applicable in the field for pest control, offering the potential to increase yields and improve farmer welfare. Further research could focus on adapting the system to a wide range of environmental conditions and scaling it up for more extensive agricultural use. The integral approach forms necessary steps toward smart farming. However, it also provides a scalable, low-cost solution for early pest detection.]]>
Copyrights © 2025
