The manufacturing line is a fundamental element in short filament fibre production, in which the gearbox is the key mechanical part. Any faults in the gearbox will greatly affect the quality of the short filament fibres. However, due to the harsh working environment, the gearbox is vulnerable to failure. Due to the complexity of the manufacturing line, effective and efficient feature extraction of gear faults is still a challenge. To this end, a new fault diagnosis method based on image recognition is proposed in this paper for gear fault detection in fibre manufacturing lines. In this method, wavelet packet bispectrum analysis (WPBA) is proposed to process the gear vibration signals. The bispectrum texture is obtained and then analysed by an image fusion algorithm for texture feature extraction. The grey-level co-occurrence matrix is used in the image fusion and the extracted texture features are four parameters of the grey-level co-occurrence matrix. Finally, a support vector machine (SVM) is adapted to recognise the gear fault type and location. Experimental data acquired from a real-world manufacturing line of short filament fibres are used to evaluate the performance of the proposed image-based gear fault detection method. The analysis results demonstrate that the newly proposed method is capable of accurate gear fault det ection in fibre manufacturing lines.