The work aims to achieve the high efficiency and high precision of LCD character detection before the packaging of electric scooters, and to solve the problems such as the difficulty of accurate segmentation and complex matching algorithm of Led segment code fonts in LCD characters. Hough line detection was conducted to correct the character region. Projection method was used to segment the character region. Morphological processing and connected domain analysis were used to extract the characters. BP neural network model was used to recognize the characters. Finally, improved geometric features were used to detect the lacking and missing lines. Gray scale features were used to detect the uneven brightness of the characters. The experimental results of LCD characters showed that the average recognition time of each character was 0.16 s and that of each screen was 0.6 s. The weighted recognition rate of LCD character defects was 96%. The algorithm has high reliability, efficiency and recognition rate, and solves the practical engineering problems of high efficiency and high precision detection of LCD characters under the defects of geometry and brightness, and provides the algorithm experience for the detection of similar products.