A Lightweight Detection Algorithm for Strawberry Ripeness

Abstract

This paper presents advancements in a streamlined maturity detection algorithm specifically tailored for strawberry harvesting, with a focus on achieving rapid and accurate identification of strawberry ripeness across challenging and varied environments.The proposed YOLOv8n-RML model incorporates several key improvements.First,YOLOv8n-RML integrates GNS modules that leverage RepConv convolution, effectively reducing the number of parameters and GFLOPS in the backbone network. Second, the adoption of the Multi-Scale Feature Pyramid Network( (MSFPN), which integrates efficient multi-scale convolution modules alongside a global heterogeneous kernel selection mechanism, significantly enhances the model feature extraction capabilities, particularly for small targets. Third, the integration of a lightweight detection head employs shared GroupNorm convolution improves localization and classification performance. Compared to the baseline YOLOv8 model, YOLOv8n-RML achieves a higher mAP of 91.3%. The YOLOv8n-RML model demonstrates a 48.6% reduction in parameters and a 34.5% decrease in GFLOPS. These improvements preserve and enhance the model’s ability to accurately detect strawberry ripeness in natural settings. Consequently, this model is highly suitable for resource-constrained embedded devices used in harvesting operations.

Abstract from the paper’s OpenAlex record; open copy licensed cc-by. Read at the source ↗

The text the model read, with the source record’s inline markup (italics, sub- and superscripts) rendered and its character entities decoded; quote offsets refer to the raw text.