Global crop pests cause substantial economic, social, and environmental damage. Accurate pest identification is crucial for effective pest management in agriculture, as different species require different control measures. However, this is a challenging task. Agricultural experts are increasingly interested in deep learning (DL) models due to their potential in image identification. Traditional methods for pest identification lack precision in recognizing and categorizing pests, mainly due to complex structure and scarcity of available data. Misidentifying pests can lead to the use of inappropriate pesticides, harming agricultural productivity and the ecosystem. This research introduces a comprehensive DeepPestNet architecture (Improved CNN) to remove this problem. The enhancement of dataset is done using image rotation techniques and evaluated DeepPestNet’s performance on new data through image augmentation techniques. The well-known Maize crops dataset was used to test the DeepPestNet structure, achieving a perfect accuracy of 100% in classifying ten pest categories. Compared with conventional pre-trained DL models, DeepPestNet performs better. This methodology provides rapid and effective support to specialists and farmers, reducing economic losses and improving agricultural yields.