Detection of maturity and counting of blueberry fruits based on attention mechanism and bi-directional feature pyramid network

The cultivation and processing of blueberries hold a significant position within the agricultural and food sectors, necessitating precise monitoring of their yield and quality. This study introduces a novel blueberry ripeness and count detection methodology that integrates an attention mechanism with a bi-directional feature pyramid network (BiFPN) within the YOLOv5 framework. The proposed attention mechanism is designed to enhance the YOLOv5 model’s focus on pertinent features while diminishing the influence of non-essential information. This is achieved by substituting the original feature fusion process in YOLOv5 with a bidirectional weighted feature pyramid structure, which facilitates more effective bidirectional feature integration, thereby augmenting the accuracy of blueberry detection.The enhanced model, designated as YOLOv5-CA, demonstrated superior performance with a mean Average Precision (mAP) at an Intersection over Union (IoU) threshold of 0.5, achieving a recall of 88.2% and a precision of 88.8%, culminating in an mAP of 91.1%. When the attention mechanism was supplemented with a bidirectional weighted feature pyramid structure, the most efficacious model, YOLOv5-SE + BiFPN, attained an mAP of 90.5%, with recall and precision rates of 88.5% and 88.4%, respectively. This configuration significantly enhanced the model’s capability to discern blueberries against complex backgrounds.Furthermore, the proposed model demonstrates a proficient detection of both the ripeness stages and the quantity of blueberry fruits, providing a foundational application for the development of automated blueberry harvesting techniques in real-world scenarios. The findings of this study lay the groundwork for future advancements in precision agriculture, particularly in the automation of fruit picking processes, by leveraging the potential of attention-based deep learning architectures.