YOLOv7-Plum: Advancing Plum Fruit Detection in Natural Environments with Deep Learning.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Plum fruit detection is crucial for yield estimation and disease monitoring in the plum industry.
• Complex orchard environments with leaf shading and fruit overlap pose significant challenges for traditional detection methods.
• Manual estimation methods are inefficient, costly, and lack accuracy.

Purpose of the Study:

• To develop an efficient and accurate plum fruit detection model for complex orchard environments.
• To improve upon existing object detection algorithms for agricultural applications.

Main Methods:

• A dataset of high-resolution plum fruit images was created from natural orchard conditions.
• An improved You Only Look Once version 7 (YOLOv7) model, termed YOLOv7-plum, was developed.
• Key modifications included incorporating the Convolutional Block Attention Module (CBAM) and Cross Stage Partial Spatial Pyramid Pooling-Fast (CSPSPPF), and using bilinear interpolation for upsampling.

Main Results:

• The YOLOv7-plum model achieved an average precision (AP) of 94.91%.
• This represents a 2.03% improvement in AP compared to the original YOLOv7 model.
• Ablation experiments and statistical analysis validated the model's effectiveness in complex backgrounds.

Conclusions:

• The proposed YOLOv7-plum model demonstrates superior performance for plum fruit detection in challenging orchard settings.
• This advancement supports the development of intelligent cultivation practices in the plum industry.
• The model offers a more accurate and efficient alternative to traditional fruit counting and monitoring methods.