The FRC materials have tremendous strength/weight ratio but are vulnerable to pull out, fracture, de-bond of fibre, fibre-matrix, and micro cracks thus, compromising tremendously with mechanical performance in different industries. It is important in detecting and classifying these defects so that the reliability and durability of these materials are guaranteed. This paper provides an automated recognition scheme based on convolutional neural networks to detect these five faults of a scanning electron microscopy image. Two hyper parameter tuners Keras Tuner and Particle Swarm Optimization were used to optimize the convolutional neural networks. The performance of Keras Tuner was 96.88 percent whereas Particle Swarm Optimization had a better result with an accuracy of 99.23 percent and a validation loss of 0.02 as compared to 0.1119 in Keras. Also, the model with Particle Swarm Optimization had greater precision (99.8 per cent), recall (99.75 per cent), and F1-score (99.77 per cent), and inference times of 25 milliseconds per batch were lower than the ones of Keras Tuner (35 milliseconds per batch). These findings demonstrate the usefulness of Particle Swarm Optimization in obtaining a greater classification accuracy. The innovation of the given research is that the developed technique integrates both modern optimization methods and convolutional neural networks to classify defects, and has surpassed the current approaches reaching an accuracy of nearly 100 percent with a much lower main complexity of computation. The results open the way to strong automated defect identification in fibre-reinforced composites, which would lead to the higher quality of materials and a guarantee of the materials performance in field.