The invention discloses a target detection method based on weak linear regression trees with incremental dimensions, which comprises the following steps: (1) preparing a training sample set (xi, yi), wherein i is equal to 1,..., N, xi expresses the characteristic values combination of training samples, yi expresses the types of samples, N expresses the number of the training samples, and N is a natural number; (2) initializing the weight of each training sample to be as shown in the specification, wherein t is a natural number and is equal to 1 in the process of initialization. (3) circularly computing the sample set for T times to acquire a linear regression tree every time to be used as a weak classifier, and then, synthesizing the T numbered weak classifiers acquired by circularly computing for T times into a strong classifier, wherein T is a natural number; and (4) classifying each region in digital images by using the strong classifier to judge whether the region is a target region or not, thereby finishing target detection. In the method of the invention, the linear regression trees are used as the weak classifiers, i.e. each weak classifier does not only contain one characteristic, but a plurality of characteristics are organically combined by the linear regression trees, thereby improving the classification capacity of the classifier; and when the trained classifier is used for detecting targets in images, higher detection rate and higher detection speed can be obtained.