Abstract: Weak and small target detection is a crucial problem in infrared imaging technology.But most of the state-of-the-art target detection methods for sequential IR images are using a single threshold for candidate extraction and accumulation schemes for weak energy justifying,this may easily cause the lost of weak and small targets and insufficient accumulation of energy for fast moving targets.Facing the above problems,a new method based on multi-level classification and backward temporal-spatial fusion is proposed.After background suppression,a step using multi-level classification is then adopted in this method for the selection of multi-class candidate targets,which enhances the capability of weak and small targets extraction.Meanwhile,according to the change of the candidate target location in adjacent frames,a kind of temporal-spatial pipeline is constructed dynamically.When candidate targets in the current frame can not be justified,a backward search is carried out for possible existence of any kinds of candidate targets through along these temporal-spatial pipelines,their energy is weighting summed with that of candidate targets in the current frame and justified with a fixed energy threshold.Finally,some experimental and comparison results using real sequential IR images is given to show the effectiveness of the new method.