Improved YOLOv8 Orchard Fruit Object Detection Algorithm Based on Enhanced Feature Fusion

Smart agriculture is becoming increasingly popular with the development of computer and artificial intelligence technology. The use of artificial intelligence technology to improve orchard management and enhance harvesting efficiency is an important technology. The current target detection algorithms based on YOLO can preliminarily complete the construction of the system. However, in these detection methods, the feature fusion part did not fully consider the actual contribution differences of different resolution feature maps to the final output, which may become a performance bottleneck when dealing with targets of different sizes and complexities. In response to the above issues, this article proposes an orchard fruit object detection algorithm based on improved YOLOv8. This algorithm improves the feature fusion part in YOLOv8 by introducing the BiFPN (Bidirectional Feature Pyramid Network) module. This method improves the accuracy and efficiency of the model in detecting objects of different sizes through a more refined feature fusion strategy. Through experiments, it has been proven that the proposed algorithm significantly improves the accuracy of object detection while ensuring computational efficiency, especially when dealing with challenging multi-scale objects. The improved algorithm improved accuracy by 16.5 percentage points, recall by 23.3 percentage points, mAP50 by 16.5 percentage points, and mAP50-95 by 20 percentage points on the avocado dataset. This research achievement has promoted the development of YOLOv8 in terms of performance in practical applications from a new perspective, and laid the foundation for future research on object detection.algorithms.