Efficient detection of usable crops from weeds is essential for effective land management and resource allocation. The challenge in earlier work towards crop-weed detection using deep learning is limited dataset and also they have not worked towards optimizing the model for improved performance. This study aims to identify the most suitable deep learning (DL) model for crop detection by evaluating the performance of separate models based on CNN, like RCNN and YOLOV8. Our DL model leverages the strengths of these architectures to achieve accurate and efficient crop detection. The results demonstrate the performance of each DL model in terms of accuracy and recall rates. This research compared RCNN and YOLOv8 deep learning models for detecting crops and weeds in protected areas. Deep learning models RCNN and Yolov8 were modified by optimizing using the traditional Bayesian and biological-inspired Algorithm, Crow Search. The Crow search algorithm has drawn inspiration from nature to enhance their performance. From the optimized Deep learning model evaluation, YOLOv8 using Crow Search emerged as the superior model, exhibiting higher precision, recall, accuracy, and IoU scores than the Optimized RCNN Model. This research signifies a significant step towards leveraging advanced technologies for land management and biodiversity conservation in protected areas. We have improved crop and weed detection accuracy by optimizing deep learning models with nature-inspired algorithms, ultimately supporting more effective land management practices. Deploying these optimized models in real-world scenarios could enhance their practical utility and contribute to sustainable conservation efforts.