Core demands
As a professional luggage inspection company, Qingdao Aibao Flower Decoration needs to inspect more than 100000 pieces of goods every day. However, the traditional X-ray needle detector can only return the internal information of the bag through X-ray, which requires human eyes to judge whether there are metal foreign matters in the bag and whether the number of metal accessories meets the production requirements, with low efficiency and high labor costs. Long time repeated work easily causes operators to use their eyes excessively and pay no attention. In addition, operators' experience level is uneven, leading to the outflow of some defective products. Therefore, it is urgent to find an early warning identification scheme with high accuracy and independent of artificial ability.
Solution
About 1000 x ray images containing needles, scissors and other foreign objects were used for Aibao Flower Ornament, and 100 x ray images of goods containing metal parts were trained using Baidu EasyDL customized image recognition. After several model iterations and verification, 14 versions were updated, and finally two models of "bag foreign object recognition" and "bag metal part recognition" that can be used on the production line were obtained. By calling the model API interface and setting different similar parameters for different styles of bags, the needle detection management system has achieved the goal of counting the number of metal parts of bags, judging whether there is any foreign matter, and sending an early warning in time. The accuracy of the obtained model is more than 90%, which effectively improves the efficiency and accuracy of needle detection.
Step 1: The bags to be inspected pass through the X-ray needle detector and take their X-ray photos
Step 2: X-ray photos are transmitted to the system background, and they are intelligently identified by calling Baidu EasyDL model API interface, and relevant parameters are returned. The server makes analysis records
Step 3: Set parameters such as pixel similarity and number of probes on the probe device side. When the API interface return parameters are inconsistent with the set parameters, an alarm will be automatically sent
Judge whether there are sewing needles in the bag
Quantity identification of metal accessories outside bags