High performing ensemble of convolutional neural networks for insect pest image detection

Pest infestation is a major cause of crop damage and lost revenues worldwide. Automatic identification of invasive insects would significantly speed up the recognition of pests and expedite their removal. In this paper, we generated ensembles of CNNs based on different topologies (EfficientNetB0, ResNet50, GoogleNet, ShuffleNet, MobileNetv2, and DenseNet201) optimized with different Adam variants for pest identification. Two new Adam algorithms for deep network optimization based on DGrad are proposed that introduce a scaling factor in the learning rate. Six CNN architectures that vary in their optimization function were trained on the Deng (SMALL), large IP102, and Xie2 (D0) pest data sets. Ensembles were compared and evaluated using several performance indicators. The best performing ensemble, which combined the CNNs using the different Adam variants, including the new ones proposed here, competed with human expert classifications on the Deng data set and achieved state of the art on all three insect data sets: 95.52% on Deng, 74.11% on IP102, and 99.81% on Xie2. Additional tests were performed on data sets for medical imagery classification that further validated the robustness and power of the proposed Adam optimization variants. All MATLAB source code is available at https://github.com/LorisNanni/.