Core demands
At present, electronic scales in the market all use traditional manual methods to query fruits, vegetables and vegetables, weigh and settle accounts, which not only requires operators to memorize the names of various vegetables and fruits, but also makes it easier to increase the waiting time of customers in line. Therefore, we hope to provide a set of fast and efficient ways for fresh food supermarkets to reduce the difficulty of operators' work with the help of AI capabilities, And save users' waiting time.
Solution
Focusing on the demand of offline supermarket for fruit and vegetable recognition, Hanyin will collect more than 20000 pictures of 150+categories and develop a fruit and vegetable recognition model on EasyDL platform using image classification capability.
After the model training, in order to better combine the model with the traditional scale, Hanyin chose the FZ3B software and hardware integration scheme of EasyDL EdgeBoard to build the AI visual scale efficiently and quickly.
When using the visual scale, the salesperson only needs to put fruits and vegetables on the scale, and the camera of the visual scale will take photos of fruits and vegetables for identification. At this time, the screen will return the results of fruit and vegetable identification. The salesperson can confirm by clicking on the screen, and the visual scale will automatically weigh and settle according to the confirmed results.
Customized image recognition solution developed based on EasyDL: allows customized development according to needs, and model training can be completed by labeling a small amount of data. When used, the advantages of this scheme are:
1. Based on EasyData, you can quickly complete data annotation and support direct import of EasyDL for training, saving data processing time;
2. EasyDL has a variety of built-in algorithm components and training templates. Based on Baidu's large-scale pre training model, high-precision models can be trained with a small amount of data;
3. Provide one-stop service of data annotation model training software and hardware integration for deployment to save development costs.