Many researchers have studied the detection of fruits in images for agricultural tasks like harvesting fruits and yield estimation. The research has been focused on overcoming the challenges of detecting and localizing fruits in the natural environment. Object detection algorithms using deep learning have effectively solved some problems related to fruit detection. The existence of varied types and species of fruits in nature is one of the problems which needs attention. The performance of deep learning methods has limitations due to the unavailability of annotated images of novel categories. Few-shot learning enables rapid adaptation to new fruit varieties without extensive retraining, addressing the challenge of evolving agricultural landscapes. N-way K-shot classification discriminates between N distinct classes with K instances per class. In this study, we have used the Eager Few-shot learning model and Resnet-50 as the backbone for fruit detection to evaluate single-class and multiclass fruits. We explore the performance of our model using various configurations by experimenting with parameters like anchor sizes, anchor scales and aspect ratios, anchor strides, and the backbone architecture to determine the optimal settings. The results demonstrate that the Eager Few-shot learning model effectively enhances fruit detection accuracy, particularly with the ResNet-50 backbone. Key findings indicate that smaller anchor strides and optimized anchor sizes contribute significantly to mean Average Precision (mAP) scores. Additionally, the study reveals that while the number of classes adversely affects detection performance, tailored configurations with optimized anchor sizes by the K-means clustering technique, yield competitive results across various metrics. These insights highlight the potential of few-shot learning methods in adapting to diverse fruit varieties.