Efficient on-tree mango detection and yield estimation using computer vision algorithms not only conserves labor and time but also aids farmers in making informed decisions regarding harvesting and marketing. Advanced machine learning approaches, while powerful, are data-intensive and may suffer from a lack of generalization due to the extensive variety of mango species, necessitating frequent retraining. This study proposes a straightforward image processing methodology for mango detection and counting, employing graph cuts and depth-first search (DFS). Initially, mangoes are detected using graph cuts without manual seeding, instead utilizing a specified color range for object identification. Following this, post-processing techniques such as thresholding and morphological erosion are applied, after which DFS is employed to count the mangoes present on the tree. Experimental results on the publicly available MangoNet semantic dataset demonstrate the efficacy of the proposed approach achieving detection and yield estimation accuracies of \(94.1\%\) and \(77.3\%\) respectively which are much higher compared to other image processing-based methods. This methodology offers a cost-effective and efficient alternative to more complex machine learning models, providing a practical solution for real-world agricultural applications.