PV panel quality control is crucial for their efficient and long-lasting operation. Detecting defects in PV panels during production is essential. Electroluminescence imaging is a commonly used method for fault detection in PV panels. This study focuses on detecting busbar slippage, a specific PV panel malfunction. Automatic error detection was researched using machine learning methods on a dataset of 500 EL images taken from the production line. Feature extraction was performed using two pre-trained deep learning architectures: ResNet and SqueezeNet. Additionally, the study aimed to observe the impact of combining features from different deep learning architectures on success parameters. The highest accuracy rate of 0.9920 was achieved using deep features extracted by Relu34 and Relu25+Conv10 layers.