Abstract—Agricultural workers want to detect, eliminate, and avoid touching small creatures such as frogs and insects in advance of and during their agricultural work. On the other side, recent researches have suggested diverse countermeasures such as developing robot arm-based machines for harvesting vegetables and pulling up weeds using camera systems; past methods have included monitoring and identifying the positive and negative targets. However, there are not sufficient previous systems for sensing and analyzing the aforementioned small creatures in farmlands. The purpose of this original research is proving the utility of our visual data analysis system based on huge image datasets using Caffe Framework for deep learning using ImageNet 2012, which connects to our program using OpenCV libraries and other outside files. In short, this study selects and executes static visual analyses using AI-based computing by tools concerning deep learning using several hidden layers after obtaining and accumulating field pictures and video data concerning small creatures such as frogs and insects in outdoor farmlands. Additionally, the author calculates the ratio between the sizes of outline of leaves on which small creatures existed as well as that of the targeted small creatures as one original standard for giving a unity to the pictures selected to some extent. Our results confirm the utility of the detection methodologies. In future, these results could contribute to the development of automatic agricultural harvesting robot-systems and to improving the daily work effectiveness of actual manual workers. Furthermore, an automatic system for eliminating small creatures could support the recruitment of agricultural workers.
 

Index Terms—picture classification using deep learning, small animals in farmlands, ImageNet 2012, Caffe Framework, OpenCV

Cite: Shinji Kawakura, "Accuracy Analyses for Detecting Small Creatures Using an OpenCV-Based System with AI for Caffe’s Deep Learning Framework," Journal of Advanced Agricultural Technologies, Vol. 6, No. 3, pp. 166-170, September 2019. Doi: 10.18178/joaat.6.3.166-170