Pest-Inspector: An Insect Detection System by Learning Data Representations

In this paper, we have introduced a beetle detection system of cultivation field of Bangladesh, which will recognize both harmful and useful insects on the basis of the real-time image. Based on this estimating analysis, the system is going to predict upcoming status and production quantities of target crops land. Our system has established on two scenarios, learning & classification and recall process. In the learning process, we have categorized two types of insect data items: Useful and Harmful and subclassified various types of harmful beetle in order to acquaint them by the help of ANN (Artificial Neural Network) terminology. Then, we have stored those classified datasets into a database repository. In next phase, the detection system will segment/cluster all of the insects from real life images and synchronize them with our pre-categorized data storage apropos to figure some educational guesses of the situation of agriculture ground and the ratio of the number of crop yields. Here, Fully Convolutional Network (FCN) had applied as a detection tool to estimates image samples. Finally, a regression technique, SVR (Support Vector Regression) has exploits to predict the desirable outcomes. An effective outcome has been presenting by through experimental study.