Predictive Model for Object Classification and Detection using Deep Learning

In the agricultural field, there are large number of objects that roam inside the field and tend to develop an unfavourable condition that may damage the crop and degrades the production. Therefore, the possibility of unavoidable situation is very high which may result into loss of human resources, agriculture assets, financial loss, and crop damage. In this paper, tiny-YOLOv3 is used to classify and detect object in real time environment, however its performance is very high, but the accuracy degrades. Thus, an enhanced model is proposed by modifying the network architecture which amplifies the real time performance, processing speed and reduces processing time. The empirical conclusion shows that the proposed model gives approximately double precision, recall, IoU, mAP, compared to actual Tiny-YOLOv3 with an improvement of 69.01%. However, the testing is performed on multiple images which also demonstrates that the proposed model gives much higher result in comparison to Tiny-YOLOv3.