An image crawler optimization method based on a convolutional nerve network comprises the following steps that 1, web crawlers are utilized to obtain initial data, the data is artificially divided into a noiseless data set Free_Noisy and a complete noise data set Pure_Noisy; 2, an initial data set is used for training a NoisyCNN network; 3, the web crawlers are utilized to search a large number of target classified image data from a main search engine and an image share website, and meanwhile the step 4 is regularly executed; 4, the NoisyCNN network is utilized to perform data screening, and the data is divided into Weak_Noisy and Pure_Noisy; 5, the extended NoisyCNN network with updated data is used; 6, whether Pure_Noisy data volume is reasonable or not is judged so as to decide to make crawling continue or not; 7, crawling is stopped. By the adoption of the image crawler optimization method, data noise can be effectively reduced, and the working efficiency can be improved.