FreqDyn-YOLO: A High-Performance Multi-Scale Feature Fusion Algorithm for Detecting Plastic Film Residues in Farmland

Mingyang Zhang ¹, Jianjie Zhang ², Yihang Peng ²

⁰College of Software, Xinjiang University, Urumqi 830091, China.

Sensors (basel, Switzerland) +

|

August 28, 2025

View abstract on PubMed

Summary

A new FreqDyn-YOLO model accurately detects plastic mulch fragments in fields, improving agricultural sustainability. This technology aids in the recovery and recycling of residual plastic film, addressing environmental challenges in farming.

Area Of Science

Agricultural Engineering
Computer Vision
Environmental Science

Background

Plastic mulch is vital for agricultural productivity but leaves residual fragments that hinder crop production and environmental sustainability.
Accurate detection and identification of residual plastic mulch fragments are technically challenging due to visual similarity with soil, irregular shapes, and variable sizes.

Purpose Of The Study

To develop an advanced detection model, FreqDyn-YOLO, for identifying residual plastic film in agricultural environments.
To enhance the discrimination between plastic mulch fragments and soil backgrounds.
To improve the detection performance across various scales and adapt to diverse film morphologies.

Main Methods

Developed FreqDyn-YOLO based on the YOLO11 architecture.
Introduced a Frequency-C3k2 (FreqC3) module with Frequency Feature Transposed Attention (FreqFTA) for enhanced feature extraction.
Implemented a High-Performance Multi-Scale Feature Pyramid Network (HPMSFPN) for effective cross-layer feature fusion.
Integrated a Dynamic Detection Head With DCNv4 (DWD4) for adaptability to varying film shapes and computational efficiency.

Main Results

FreqDyn-YOLO achieved a 5.37% improvement in mAP50, 1.97% in precision, and 2.96% in recall compared to the baseline approach on a custom agricultural dataset.
The model demonstrated superior performance against other recent detection methods.
Experimental results validated the model's effectiveness in identifying residual plastic film.

Conclusions

FreqDyn-YOLO provides a robust technical foundation for precise residual plastic film identification in agriculture.
The developed model shows significant promise for integration into automated systems for residual mulch recovery and recycling.
This research contributes to sustainable agricultural practices by addressing the challenges of plastic mulch residue.

Keywords:

DCNv4 FreqFTA HPMSFPN YOLO11 residual film