HierbaNetV1: A novel convolutional neural network architecture

Abstract

Abstract Artificial intelligence also referred to as AI is growing at an exponential rate in diverse fields. Convolutional Neural Networks (CNN), have become largely established and are essential to the advancement of AI. We suggest HierbaNetV1, a novel neural network architecture with a cutting-edge feature extraction method. This architecture extracts and integrates both low-level and high-level features. The high-level features that are generated exhibit features with different levels of complexity. The resulting low-level features show the essential traits of the region of interest, which. Feature integration facilitates a deeper comprehension of the object's patterns. The architecture is made up of 72 layers with 14.3 million parameters. The HierbaNetV1 model is constructed using the crop-weed dataset SorghumWeedDataset_Classification. After being trained, verified, and tested on the dataset, the model produces an overall classification accuracy of 98.6%, proving that HierbaNetV1 is a state-of-the-art architecture. This article's accompanying video tutorial demonstrates a thorough comprehension of HierbaNetV1's distinct feature extraction procedure. The upcoming enhancements are also included in the summary that is provided in the video tutorial.