This research paper explores the application of reversible neural networks (RevNNs) in the domain of weed classification, aiming to enhance the accuracy and efficiency of automated weed species identification. Leveraging the unique bidirectional information flow inherent in RevNNs, this study presents a comprehensive methodology involving feature extraction, network architecture design, and training for discerning complex patterns within weed images. To construct a discerning feature set, distinct characteristics are extracted from the weed imagery employing methods such as structural parameters, texture analysis, and morphological attributes. These features, embodying essential information regarding weed species, undergo aggregation and refinement before being inputted into the Reversible Neural Network architecture. The RevNN processes this feature ensemble to discern and classify various weed species or characteristics. In the course of empirical investigation and validation conducted on two distinct datasets, namely the plant seedling dataset and DeepWeeds, the performance of the proposed RevNN model in the accurate classification of diverse weed species is systematically evaluated. The experimental evaluation yields an overarching accuracy of 99.08% and 97% on the respective test datasets of plant seedling and DeepWeeds datasets. The findings demonstrate the potential of RevNNs in robustly identifying different weed species, offering promising implications for advancing agricultural practices and weed management.