A novel insect and pest identification model based on a weighted multipath convolutional neural network and generative adversarial network
Vinita Abhishek Gupta, M.V. Padmavati, Ravi R Saxena, Raunak Kumar Tamrakar
Abstract
Timely identification of insects and their management play a significant role in sustainable agriculture development. The proposed hybrid model integrates a weighted multipath convolutional neural network and generative adversarial network to identify insects efficiently. To address the shortcomings of single-path networks, this novel model takes input from numerous iterations of the same image to learn more specific features. To avoid redundancy produced due to multipath, weights have been assigned to each path. For Xie2 dataset, the model shows 3.75%, 2.74%, 1.54%, 1.76%, 1.76%, 2.74 %, and 2.14% performance improvement from AlexNet, ResNet50, ResNet101, GoogleNet, VGG-16, VGG-19, and simple CNN respectively. To the best of our knowledge, no researchers have used a multipath convolution neural network in insect identification.