BeetleID: An Android Solution to Detect Ladybird Beetles

Abstract

In this work, an Android mobile application named BeetleID was developed to detect ladybird beetles through image pre-processing methods and a deep learning convolutional neural network model. The image pre-processing module consists of three main algorithms: saliency map, active contour, and superpixel segmentation. The used convolutional neural network was validated with a 2611 image set of ladybird beetle species with a five-fold cross-validation method. It achieved accuracy and area under the curve of the receiver operating characteristic scores of 0.92 and 0.98, respectively. Furthermore, the application's feasibility was assessed by the mean execution time and battery consumption metrics of mobile emulators, phone Pixel 3a XL and tablet Pixel C, which obtained 16.32 and 18.43 seconds 0.07 and 0.11 milliampere-hour, respectively. These results prove that the proposed application is an excellent solution, with a few optimization issues, for specialists to detect ladybird beetles in wildlife environments accurately.