Agriculture is one of the most important industries contributing to several countries’ national income. Timely identification of these pests is crucial for both farmers and agriculture experts to implement effective mitigation methods. Traditionally, farmers employ manual techniques based on their experience and visual inspection to identify paddy pests, but this is highly inefficient, time-consuming, and error-prone. At times, even experienced farmers and agriculture experts might struggle to identify a specific species of pests due to their size, colors similar to leaves, and the extensive variety of identical types. Moreover, farmers often apply a large quantity of pesticide without accurately identifying the exact type of pests and their underlying causes. Therefore, it is increasingly important to automate the process of detection of paddy pests to reduce pesticide usage and subsequently minimize the loss in yield. In this paper, we present an open dataset and deep learning models for automated pest identification in real paddy fields. Our dataset contains 6,062 annotated paddy leaf images across 17 classes (16 pest classes and a normal class). We benchmarked our dataset using six pre-trained models (ResNet34, VGG16, DenseNet121, EfficientNetv2m, MobileNetV3Large, and SqueezeNet1_0). The experimental results showed that ResNet34 achieved the highest accuracy of 98.31%. We release our dataset and reproducible code in the open source for community use.