DEVELOPMENT OF A PEST MANAGEMENT SYSTEM FOR PRECISION AGRICULTURE USING TRANSFER LEARNING AND INTERNET OF THINGS (IOT)

The aim of this study is design and implementation of pest management system for precision agriculture using integrated transfer learning and internet of things technique. The methodology used is dynamic system development model. The tested for primary data collection is Aninri and the pest considered are weevil, caterpillar, and whitefly. The secondary data source is Kaggle. Then total sample size of data collected is 18138. The data was processed through annotation and labeling using Robowflow tool. The benchmark transfer learning model used is You Only Look Once(YOLOv10), which was improved through the optimization of the spatial pyramid pooling function using average pooling layer and maximum pooling layer. Python programming language was applied to train the model and generate the real-time pest classification solution. Training of the model was done experimentally, considering the benchmark model and then the improved YOLOv10. From the results, it was observed that precision of the model with YOLOv10 recorded 0.88519 while the recall recorded 0.84463. The mAP@0.5 recorded 0.82746 and 0.58272 for the traditional YOLOV-10, while for the new model the precision recorded 0.97519, recall reported 0.93163 and mAP@0.5 recorded 0.92486 and 0.67272 respectively. From these results, it was observed that the new YOLOV-10 in all metrics supersedes the traditional models and the reason was due to the optimization of the SPPF layer which increased the quality and quantity of features extracted, thereby optimizing the model performance in correctly pest detection in the farm.