Application of Deep Learning-Based Multi-Scale Feature Fusion in the Visual System of Precision Welding Robots
Bibliographic record
Abstract
In high-end equipment manufacturing and aerospace industries, the quality of precision welding directly affects product reliability.However, the welding process is often challenged by complex lighting variations, metal spatter, torch occlusion, and multi-scale defect characteristics, which pose significant difficulties for defect detection in robotic visual systems in terms of both accuracy and real-time performance.Traditional handcrafted feature methods and early deep learning models suffer from insufficient utilization of multiscale features and inadequate fusion of contextual semantics, resulting in high missed detection rates of small defects and failures in occluded scenarios.Existing single-scale feature networks tend to overlook low-level detail information, and conventional feature fusion methods fail to fully exploit cross-resolution feature complementarity.In addition, fixed anchor box schemes lead to high localization errors, and the lack of online compensation mechanisms for dynamic occlusions hinders detection performance in realworld applications.To address these challenges, this paper proposes a real-time welding defect detection method based on multi-resolution feature fusion tailored for the visual system of precision welding robots.The research encompasses six key aspects: data acquisition, optimization of the detection network, backbone network enhancement, multilayer feature fusion, adaptive anchor box adjustment, and occlusion-aware stereo vision measurement.By constructing a diverse multi-condition dataset, introducing cross-layer attention mechanisms, and designing an adaptive feature fusion strategy along with a spatiotemporal joint compensation model, the proposed method effectively overcomes the limitations of single-scale feature dependence.Experimental results demonstrate significantly improved detection accuracy for multi-scale defects under complex conditions and enhanced adaptability in dynamic scenes.The outcomes of this study offer a reusable technical framework for industrial visual inspection and provide meaningful contributions toward the intelligent development of precision welding.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".