In evaluating modern architecture, crop pest detection becomes a challenge that later impacts yield and sustainability. To ensure food security and minimize losses, Effective pest detection in very essential in crops. This research proposes the InceptionV3 architecture, an enhanced deep learning model, and augments it with custom top layers to improve feature extraction and classification accuracy. The architecture incorporates a series of dense layers with ReLU activation, L2 regularization, batch normalization, and dropout to mitigate overfitting. We have used 256 \(\times \) 256 RGB images for training the peat dataset using the Adam optimizer with a learning rate of 0.0002. Our experimental results demonstrated a training accuracy of 100% and test accuracy of 91.45%, which outperformed our other suggested architecture, DenseNet201, that having 88.0% test accuracy. Our dataset is vast, with 5494 images carrying 12 classes of agricultural pests. Compared to some recent models, InceptionV3 offers a robust balance between accuracy and generalization that makes it well-suited for real-world agricultural applications. By ensuring a high-performance framework for agriculture, the researchers were able to advance automated pest detection with potential applications in integrated pest management systems.