Cotton boll localization method based on point annotation and multi-scale fusion.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Precise cotton boll localization is crucial for intelligent cotton field management, including automated picking and yield estimation.
• Current object detection methods rely on expensive bounding box annotations, which can include non-object regions, potentially leading to model misjudgments.
• Point annotations offer a cost-effective alternative with higher object specificity.

Purpose of the Study:

• To propose a novel point annotation-based multi-scale cotton boll localization method, MCBLNet.
• To address the limitations of traditional object detection methods in terms of annotation cost and accuracy.
• To enhance the precision and robustness of cotton boll localization for agricultural applications.

Main Methods:

• MCBLNet utilizes point annotations for training, reducing labeling costs and improving feature relevance.
• The method comprises scene encoding for feature extraction, location decoding for localization prediction, and localization map fusion for multi-scale information integration.
• Experiments were conducted on a custom Cotton Boll Localization (CBL) dataset comprising 300 in-field images.

Main Results:

• MCBLNet demonstrated a significant improvement in average precision, achieving a 49.4% increase on the CBL dataset compared to existing point-based localization methods.
• The proposed method showed performance comparable to or exceeding common object detection techniques.
• Experimental results validate the robustness and accuracy of MCBLNet for cotton boll localization.

Conclusions:

• MCBLNet offers a more efficient and accurate solution for cotton boll localization by leveraging point annotations.
• The multi-scale approach effectively fuses information, enhancing localization performance.
• This method provides a valuable tool for advancing intelligent cotton production and management systems.