Cross-domain discrepancy frequently hinders deep neural networks to generalize to different datasets. In order to improve the robustness of shadow detection, a novel unsupervised domain adaptive shadow detection framework was proposed. Firstly, in order to reduce the data bias between different domains, a multi-level domain adaptive model was introduced to align the feature distributions of source domain and target domain from low level to high level. Secondly, to improve the model ability of soft shadow detection, a boundary-driven adversarial branch was proposed to guarantee the structured shadow boundary was also able to be obtained by the model on the target dataset. Thirdly, the entropy adversarial branch was combined to further suppress the high uncertainty at shadow boundary of the prediction result, so as to obtain an accurate and smooth shadow mask. Compared with the existing deep learning-based shadow detection methods, the proposed method has the Balance Error Rate (BER) averagely reduced by 10.5% and 18.75% on ISTD dataset and SBU dataset respectively. The experimental results demonstrate that the shadow detection results of the proposed algorithm have better boundary structure.

Aiming at the problems that a wide variety of Chinese medicinal materials have small samples, and it is difficult to classify the vessels of them, an improved convolutional neural network method was proposed based on multi-channel color space and attention mechanism model. Firstly, the multi-channel color space was used to merge the RGB color space with other color spaces into 6 channels as the network input, so that the network was able to learn the characteristic information such as brightness, hue and saturation to make up for the insufficient samples. Secondly, the attention mechanism model was added to the network, in which the two pooling layers were connected tightly by the channel attention model, and the multi-scale cavity convolutions were combined by the spatial attention model, so that the network focused on the key feature information in the small samples. Aiming at 8 774 vessel images of 34 samples collected from Chinese medicinal materials, the experimental results show that by using the multi-channel color space and attention mechanism model method, compared with the original ResNet network, the accuracy is increased by 1.8 percentage points and 3.1 percentage points respectively, and the combination of the two methods increases accuracy by 4.1 percentage points. It can be seen that the proposed method greatly improves the accuracy of small-sample classification.

Aiming at the problem that the security surveillance cameras have been hidden by leaves, a leaf occlusion detection algorithm based on Support Vector Machine (SVM) was proposed. The algorithm contains three steps. First, the regions of the leaf existing in the video were segmented. The accumulated frame subtraction method was applied to achieve this purpose. Second, the color and area information of the whole video image and the segmented regions were extracted as the key features. Third, these features were used for modeling and detecting obstacle occlusion by SVM. For all the collected samples, the detection accuracy of this method can reach up to 84%. The experimental results show that the proposed algorithm can detect the leaf occlusion in security surveillance video effectively.

A new method based on watershed algorithm was proposed to raise the segmentation accuracy of the crop disease leaf images. At first, distance transformation and watershed segmentation were conducted on the binary crop disease leaf images to get the background marker, and the preliminary foreground markers were generated by extracting the regional minimum from the reconstructed gradient images, and then some fake foreground markers were eliminated by the further filter. In the next step, both background markers and foreground markers were imposed on the gradient image by the compulsive minimum algorithm. At last, the watershed transformation was carried out on the modified gradient image. Lots of cucumber disease leaf images were segmented effectively using the method. The results of experiment indicate that disease spots can be separated precisely from the crop leaf images. Additionally, the segmentation results are not influenced by leaf texture and its accuracy is up to more than 90 percent, so the method has certain validity and practical value.