Weeds, which compete with other plants for water and sunlight, negatively impact crops and reduce crop output. Identifying weeds in their early stages of growth is essential to reduce their influence on crop development and increase production. This study aims to classify crops and the weeds that frequently grow alongside them. The paper examine various models, including SVM, KNN, Decision Tree, Random Forest, and CNN, and the results show that CNN has a higher accuracy than the other approaches. Thus, this study proposes a method combining CNN and PFO to improve weed detection performance. CNN is used to identify complex pattern dependencies in the images, and the Peregrine Falcon Optimizer (PFO) is used for hyperparameters optimization. An experimental result shows that CNN-PFO model outperformed the other optimizers, including PSO, GA, GGO, and GWO. The performance metrics used for the evaluation are accuracy, sensitivity (recall), specificity, F1-score, and Positive Predictive Value (PPV). The test results also show a notable increase with precision of 99.5, recall of 99.3 and accuracy of 99.4%.