A possible new strategy for detecting internal and subtle defects in fruits using artificial intelligence

The fruit production industry is constantly faced with the challenge of classifying and selecting quality
products. Defects in fruit can be both internal and external, and detecting these defects is crucial to
ensuring product quality and safety. We are working on a possible new approach to fruit defect detection
using our AI systems.

IngiGrader is an advanced artificial vision system that has demonstrated exceptional performance in
detecting and classifying defects in fruit. During testing, it was observed that the AI was able to detect and
classify defects that were not directly visible in the image. In some cases, these defects were internal or
external but very subtle. We suspect that the AI detects these defects based on characteristics that may not
be directly related to the defect itself. Below, we present how we plan to develop this discovery to
improve classification systems.

The methodology used in our artificial vision system is based on the combination of deep learning
techniques, AI, and neural networks. The process of detecting and classifying defects in fruit is carried out
through the following sequence of steps:

1. Image acquisition: High-resolution images of the fruit are captured using state-of-the-art cameras and controlled lighting systems.
2. Preprocessing: Captured images are subjected to cleaning and enhancement processes, such as noise removal, lighting correction, and image normalization.
3. Feature extraction: Relevant features are extracted from preprocessed images using image processing and machine learning techniques.
4. Neural network training: A deep neural network is trained using a labeled dataset that includes images of fruit with and without defects. The network learns to recognize patterns and features related to the presence of defects in fruit.
5. Defect classification: Finally, the neural network classifies fruit images according to the presence or
   absence of defects and their severity.

Discovery of Non-Visible and Subtle Defects

During testing of our vision system, defects were detected and classified that were not directly visible in
the image. Although we initially considered these to be AI errors, we now suspect that the model detects
these defects based on characteristics that may not be directly related to the defect itself but rather
indirectly.

This suggests that the neural network has learned to identify underlying patterns and
characteristics that are not evident to the human eye or traditional image processing approaches.
We believe that this discovery could have great potential to improve defect detection and classification
systems. By exploiting the ability of AI to detect non-evident characteristics, it is possible to develop
more accurate and efficient classification systems. This results in better fruit selection, which in turn leads
to greater customer satisfaction and minimizes waste of quality products.

To further develop this strategy, our company plans to conduct additional research and develop the
necessary technology. The next steps include:

• In-depth analysis of characteristics: A comprehensive analysis of the characteristics detected by the AI will be conducted to better understand how they relate to defects and how they can be exploited to improve classification.
• Neural network optimization: A neural network optimization process will be carried out, adjusting
  hyperparameters and architecture to further improve the accuracy and ability to detect non-visible and subtle defects.
• Development and validation of an improved grading system: Based on the findings of the analysis
  and optimization, an improved classification system will be developed that incorporates the ability to
  detect and classify non-visible and subtle defects. This system will be validated using additional datasets and tested in real production environments.
• Integration and adaptation to different types of fruit: The improved system will be adapted and
  integrated to work with different types of fruit and in various production conditions. This will allow fruit producers to benefit from the technology in a wide variety of applications and environments.
• Commercialization: Once the improved system has been validated and optimized, we will incorporate the new decision models developed into the core of IngiGrader.
  
  We will keep you informed about this new project!