The invention relates to the technical field of security and protection monitoring, in particular to a method and
system for detecting abnormal work of a
network camera, and the method comprises the steps: training a basic
data set to form a
large model, collecting normal and abnormal pictures, carrying out the blocking
processing, labeling, preprocessing, and carrying out the multi-
task learning training of a CNN
network structure model. SSIM is used as a
loss function, simulated abnormal data is generated to enhance generalization ability, and model performance is evaluated; after video
stream key frames or fragment blocks are obtained, a result R0 is recognized through the
large model, reviews regularly, if abnormity occurs, the result is output, and after a certain result is accumulated, the
large model is trained. The
system comprises a
data acquisition module, a model training module, a detection
data acquisition module, a
data partitioning module, an
anomaly detection module, an examination module, a model optimization module and the like. According to the method, the detection efficiency and accuracy are improved, manual intervention is reduced, the problems of low video multi-channel
detection performance, tedious threshold setting and low detection accuracy in the prior art are effectively solved, and the method is suitable for various security and protection monitoring scenes.