Targeted application of deep learning to automated visual inspection equipment
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- AMGEN INC
- Filing Date
- 2025-03-27
- Publication Date
- 2026-06-09
AI Technical Summary
【0007】 AVIステーション又は外部の処理構成要素は、2次元画像のピクセル値(例えば、正規化されたピクセル強度値)を訓練済みニューラルネットワークに提供する。これにより、容器サンプルが受理不可能であるか(例えば、画像化されたエリア内の粒子が、受理不可能な数、サイズ、及び/又はタイプを含有するか)否かが推論される。ニューラルネットワークは、例えば、粒子及び/又はガス充填気泡の受理可能又は受理不可能な数、タイプ、サイズなどを有することが知られている(及びラベル付けされている)サンプルの幅広い2次元画像を使用して、教師あり学習手法で訓練されてもよい。ニューラルネットワークの訓練するために使用される画像の選択及び分類は、推論フェーズにおける性能にとって重要である。更に、欠陥のあるユニットを受理することを回避するために、予期しない状況を予測し、それを訓練画像に含めるべきである。重要なことに、訓練されたニューラルネットワーク、又はニューラルネットワークを含むより大きな推論モデルは、再認定なしにモデルを変更すること(例えば、更に訓練すること)ができないように、認定前に「ロック」されてもよい。システムが手動の目視検査と同等に又はそれよりも良好に機能することを確実にするために、好ましくは、受理基準が確立され事前に承認されているべきである。
Smart Images
Figure 0007872400000006 
Figure 0007872400000007 
Figure 0007872400000008
Abstract
Claims
1. A method for improving the accuracy and efficiency of automated visual inspection of containers, Orienting a line scan camera over a container containing a sample via one or more processors, such that the line scan camera is angled upward with respect to the horizontal plane so as to match or approximate the inclination of the container's stopper. The container is spun via one or more processors, The process involves spinning the container while capturing multiple images of the stopper's edge using the line scan camera, wherein each of the multiple images corresponds to a different rotational position of the container. The one or more processors generate a two-dimensional image of the edge of the stopper based on at least the plurality of images, One or more processors that execute an inference model including a trained neural network process the pixels of the two-dimensional image to generate output data indicating at least one of (1) the likelihood that the sample contains a defect, and (2) whether the sample is acceptable. A method that includes this.
2. The method according to claim 1, further comprising using one or more processors to ensure that the container is selectively transported to a designated rejection area based on the output data.
3. The method according to claim 1, wherein processing the pixels of the two-dimensional image includes applying intensity values associated with different pixels, or other values derived from the intensity values, to different nodes in the input layer of the trained neural network.
4. The method according to claim 1, wherein the container is a syringe and the stopper is a plunger.
5. The method according to claim 1, further comprising transporting the container using an electric rotary table or star wheel in response to a command generated by one or more processors.
6. The method according to claim 1, further comprising inverting the container in response to a command generated by one or more processors such that the stopper is below the sample.
7. The method according to claim 1, wherein spinning the container includes rotating the container at least 360 degrees about the central axis of the container in response to a command generated by one or more processors.
8. The line scan camera is a first line scan camera, the plurality of images are a first plurality of images, the container is a first container, the two-dimensional image is a first two-dimensional image, and the method is While the first line scan camera is being directed, the second line scan camera is also directed relative to the second container via the one or more processors so that it is angled upward with respect to the horizontal plane so that it matches or approximates the inclination of the stopper of the second container. The first container is spun while the second container is spun via one or more processors, While capturing the first plurality of images, and while the second container is spinning, the second line scan camera captures a second plurality of images of the stopper of the second container, wherein each of the second plurality of images corresponds to a different rotational position of the second container. A second two-dimensional image is generated based on at least the second plurality of images via one or more processors. The method according to claim 1, further comprising:
9. The method according to claim 1, further comprising training the neural network via the one or more processors using a labeled two-dimensional image of the stopper edge of the container before processing the pixels of the two-dimensional image.
10. The method according to claim 9, comprising training the neural network using labeled two-dimensional images of containers containing samples, including different types, numbers, sizes, and locations of objects.
11. An automated visual inspection system, Line scan camera and Positioning hardware configured to orient the line scan camera over a container containing a sample, to angle the line scan camera upward with respect to the horizontal plane so that it aligns with or approximates the inclination of the container's stopper, and to spin the container while it is oriented in this manner; A memory for storing instructions, wherein, when an instruction is executed by one or more processors, the memory causes the one or more processors to execute the method according to any one of claims 1 to 4, 6, 7, 9, or 10. An automated visual inspection system equipped with the following features.
12. The automatic visual inspection system according to claim 11, wherein the positioning hardware includes an electric rotary table or a star wheel, and the container is oriented by transporting the container using the electric rotary table or the star wheel.
13. The line scan camera is the first line scan camera, the plurality of images are the first plurality of images, the container is the first container, the positioning hardware is the first positioning hardware, the two-dimensional image is the first two-dimensional image, and the output data is the first output data. The automated visual inspection system according to claim 11, further comprising a second line scan camera and a second positioning hardware configured to orient the second line scan camera with respect to a second container containing a second sample, such that the second line scan camera is angled upward with respect to the horizontal plane so that it aligns with or approximates the inclination of a stopper of the second container, and to spin the second container while it is oriented in this manner.