As an intelligent algorithm software platform with deep learning algorithm and industrial vision technology as the core, Handdle AI has reached the industry leading level in small sample learning, transfer learning, visual debugging, etc.; the built-in 2D and 3D function modules can realize defect detection under complex backgrounds and solve industrial detection problems.


Strengthen the code reading algorithm, improve the recognition rate, and optimize the scanning scenarios such as deformed and fuzzy codes

Code reading
3D detection

Mark and train the 2D images converted from 3D point cloud data to generate AI neural network model

Dimensional measurement

Realize accurate calculation in the image by acquiring pixel information in the image


Carry out the pixel level detection for the tested materials, such as identification of silicon wafer crack area, impact bearing area, etc.

Core Functions

Classify and position the detected materials. Applicable to detection, counting, etc. of the multiple/small objects, including drug pill counting, 3C electronic component detection and other scenarios


Conduct the training through positive samples and use model reasoning to predict the type of new pictures

One class training

Tag and recognize single character and multiple characters.

Characters under different backgrounds can be recognized, for example, the character information of products or components in various complex scenarios, such as steel seals and laser engravings


Classify and judge the tested materials, such as OK/NG classification of materials, color and type of tested objects, and fine classification of 3C defects

Minimum sample size <=5
Fully accelerate the optimization of software and hardware. On the basis of ensuring the detection accuracy and precision, it can detect small articles much faster than the traditional algorithms.
Hardware accelerating optimization
Stability of highly integrated machine learning and generalization ability of deep learning
Machine learning & deep learning
Taking advantage of the machine learning algorithm's application in detection speed and detection stability and in combination with the characteristics of deep learning algorithm such as interference, efficient feature extraction and generalization ability, it can effectively enhance the item detection stability and detection accuracy 
Single positive sample
Strengthen the model optimization direction and reduce the optimization time
Solve the black box problem generated by the neural network under the module "Classification" of deep learning algorithm, and guide the optimization of neural network model through visual thermodynamic diagram
Visual debugging
Enhance the universality of neural network model
Based on the extraction and training of common features of products made of the same material, improve the detection universality of neural network model, transfer the model to more product models and solve the application scenario problems of fewer samples and more models
Transfer learning
Effectively improve the detection accuracy by multi-dimensional feature extraction
Conduct the multi-dimensional feature extraction for training set images, auto-filter the effective feature information and increase the training weight of such data to effectively improve detection accuracy
Image feature enhancement
Save NG samples and realize the rapid project deployment

Enhanced detection of pixel coincidence

Reduce the excessive dependence of AI algorithm on data size
In the detection of common types of defects, it can realize quick training and deployment on the basis of a small amount of data, winning rather high detection accuracy
Small sample training
Precise feature recognition
By optimizing the underlying framework of the algorithm, the detected target area is infinitely coincident with the detection result, so as to improve the detection accuracy and greatly reduce the rate of over detection and missed detection

Core Technology Advantages

For products with clear image features and single defect type, high detection accuracy can be achieved in a short time through OK sample training, so as to achieve the rapid project deployment
Number of compatible models30
From image data preprocessing, to algorithm and neural network optimization, and then to underlying algorithm optimization

Deployment Process

If you have relevant business needs, please leave a message

Product Consulting

Leave a message