A manufacturer wants a model that inspects photos of circuit boards on a production line and flags visible solder defects. Which neural network architecture is best suited to this task?
Show answer and explanation
Answer: D
Convolutional neural networks apply learned filters across the spatial grid of an image, picking up local patterns such as edges, shapes, and texture that indicate a solder defect. This makes CNNs the standard architecture for image classification and defect detection. An LSTM is built for ordered sequences, a GAN is mainly used to generate synthetic samples, and a sequence autoencoder targets ordered data rather than spatial image classification.
Why the other options are wrong
- A. LSTMs process ordered sequences such as text or sensor time series. They are not the usual choice for classifying regions of a still photograph.
- B. A sequence autoencoder is designed to compress and rebuild ordered data. It is not the standard choice for labeling defects in a photo.
- C. GANs generate new synthetic samples. The task here is to classify real photos, which a GAN does not do directly.