Towards Automatic Power Battery Detection: New Challenge, Benchmark Dataset and Baseline (CVPR 2024)

Xiaoqi Zhao1,2*, Youwei Pang1,2*, Zhenyu Chen1, Qian Yu1,
Lihe Zhang1*, Hanqi Liu2, Jiaming Zuo2*, Huchuan Lu1,
1IIAU-Lab, Dalian University of Technology,
2AI Lab, X3000 Inspection Co., Ltd,
Paper Dataset&Code

If you are interested in Ai4Industury, feel free to contact with us via Email (zxq@mail.dlut.edu.cn, lartpang@mail.dlut.edu.cn),

Application Value

empty

Assembly process of power battery for new energy vehicles. The power battery has gone through the process from the cell to the system before it is finally installed on the vehicle unit. To ensure the safety of the power battery, it is necessary to perform power battery detection (PBD) on each battery cell to complete its functional evaluation.

Task Definition

empty

Power Battery Detection (PBD) aims to judge whether the battery cell is OK or NG based on the number and overhang. Therefore, object counting and localization are necessary processing for PBD, which can provide accurate coordinate information for all anode and cathode endpoints.

Abstract

We conduct a comprehensive study on a new task named power battery detection (PBD), which aims to localize the dense cathode and anode plates endpoints from X-ray images to evaluate the quality of power batteries. Existing manufacturers usually rely on human eye observation to complete PBD, which makes it difficult to balance the accuracy and efficiency of detection. To address this issue and drive more attention into this meaningful task, we first elaborately collect a dataset, called X-ray PBD, which has 1,500 diverse X-ray images selected from thousands of power batteries of 5 manufacturers, with 7 different visual interference. Then, we propose a novel segmentation-based solution for PBD, termed multi-dimensional collaborative network (MDCNet). With the help of line and counting predictors, the representation of the point segmentation branch can be improved at both semantic and detail aspects. Besides, we design an effective distance-adaptive mask generation strategy, which can alleviate the visual challenge caused by the inconsistent distribution density of plates to provide MDCNet with stable supervision. Without any bells and whistles, our segmentation-based MDCNet consistently outperforms various other corner detection, crowd counting and general/tiny object detection-based solutions, making it a strong baseline that can help facilitate future research in PBD. Finally, we share some potential difficulties and works for future researches. The source code and datasets will be publicly available at \href{https://github.com/Xiaoqi-Zhao-DLUT/X-ray-PBD}{X-ray PBD}.

Dataset

empty

Statistics of the X-ray PBD dataset. (a) Taxonomic of interference and shots. (b) Overhang distributions. (c) Number distributions. (d) Co-occurrence distribution of attributes. (e) Multi-dependencies among these attributes.

empty

Examples of various attributes from our X-ray PBD dataset (best viewed zoomed in).

empty

Attribute descriptions.

Research Value&Challenges

(1) Open vision modeling problem.

(2) Minimal object instance localization (1 pixel/X,000,000+ pixel).

(3) Weak feature perception.

Method: MDCNet

empty

Overview of our MDCNet. It contains a shared encoder to extract different level features for the prompt and current images, respectively. Multi-scale module is only embedded in the high-level features. Prompt filter module are used to combine the prompt and current features to generate a series of filtered features. Point predictor include five decoder layers to produce point segmentation maps. Counting and line predictors are guided by the point prediction and fusing high-level and low-level features, respectively.

Comparison with Potential Solutions

empty

Quantitative comparison of different methods. ``—'' represents that the results are not available because these methods can not provide coordinate information or their prediction accuracy of the number of plates is zero.

empty

Qualitative results on the regular, difficult and tough examples with different shots and attributes. Best viewed zoomed in.

empty

Visual comparison with other general/tiny object detection-based, counting-based, corner detection-based solutions. We directly visualize the predicted results (Ours: Segmentation map, General/Tiny object detection methods: Bounding box, Counting methods: Density map, Corner detection methods: Corner map) without any post-processing operations.

Coming Soon

empty

Ai4Industury-Image Blind Enhancement.

empty

Ai4Industury-CT Reconstruction.

empty

Ai4Industury-Multimodal Unified Model.

empty

Ai4Industury-GPT.

BibTeX


  @inproceedings{X-ray-PBD,
  title={Towards Automatic Power Battery Detection: New Challenge, Benchmark Dataset and Baseline},
  author={Zhao, Xiaoqi and Pang, Youwei and Chen, Zhenyu and Yu, Qian and Zhang, Lihe and Liu, Hanqi and Zuo, Jiaming and Lu, Huchua},
  booktitle={CVPR},
  year={2024}
}
}