Segmentation of land cover in remote sensing images is a task that involves interpreting remote sensing data using machine vision. Satisfying segmentation results in agriculture and forestry regions can guide land resource management, natural environment protection, urban construction, and the distribution of agricultural products. However, the performance of the widely used deep learning segmentation model on high-resolution remote sensing segmentation datasets in agriculture and forestry regions needs to be improved. To solve the problems of poor accuracy and loss of context information in remote sensing image semantic segmentation, this paper proposes an improved semantic segmentation network architecture. The model utilizes multi-scale feature extraction, deploys a multi-layer attention feature fusion module and an up-sampling fusion module to capture high-quality multi-scale context information, correctly handle scale changes, and help narrow the semantic gap between different levels. Finally, the proposed MLP decoder refers to the dynamic up-sampling operator to aggregate the information at different levels to achieve pixel segmentation. To verify the effectiveness of our proposed model, the researchers conducted experiments on two land cover segmentation datasets. The training process specifically designs data augmentation strategies for remote sensing segmentation tasks to enhance the model’s generalization ability. The final model achieved an mIoU (mean Intersection over Union) of 65.05% on the self-built rural land cover datasets, surpassing the benchmark network UPerNet by 5.92%. On the LoveDA dataset, our model achieved state-of-the-art performance with an mIoU of 53.39%, demonstrating its versatility.

Comparative analysis of quality and safety monitoring technology has always been the most sensitive topic in the food field. Quality and safety monitoring of liquor is the prerequisite for maintaining market stability and guaranteeing quality. Therefore, it is very important to judge the quality of liquor correctly. In this study, the XGBoost model was used to evaluate the liquor quality of 4899 liquor samples from the Vinho Verde white wine sample dataset in northern Portugal. The influence of 11 liquor characteristics such as volatile Acidity, alcohol, density, and free sulfur dioxide on liquor quality was explained by SHAP, PDP, and other interpretation techniques. The results showed that the XGBoost model had better evaluation performance than the general model. Among the 11 liquor characteristics studied in this paper, liquor quality was affected not only by alcohol content but also by volatile acid, density, and residual sugar. This study can not only provide an objective evaluation method for liquor quality but also make people a further intuitive understanding of the evaluation results, so as to produce better liquor quality.

Tumor metastases detection is of great importance for the treatment of breast cancer patients. Various CNN (convolutional neural network) based methods get excellent performance in object detection/segmentation. However, the detection of metastases in hematoxylin and eosin (H&E) stained whole-slide images (WSI) is still challenging mainly due to two aspects. (1) The resolution of the image is too large. (2) lacking labeled training data. Whole-slide images generally stored in a multi-resolution structure with multiple downsampled tiles. It is difficult to feed the whole image into memory without compression. Moreover, labeling images for the pathologists are time-consuming and expensive. In this paper, we study the problem of detecting breast cancer metastases in the pathological image on patch level. To address the abovementioned challenges, we propose a few-shot learning method to classify whether an image patch contains tumor cells. Specifically, we propose a patch-level unsupervised cell ranking approach, which only relies on images with limited labels. The main idea of the proposed method is that when cropping a patch A from the WSI and further cropping a sub-patch B from A, the cell number of A is always larger than that of B. Based on this observation, we make use of the unlabeled images to learn the ranking information of cell counting to extract the abstract features. Experimental results show that our method is effective to improve the patch-level classification accuracy, compared to the traditional supervised method. The source code is publicly available at https://github.com/fewshot-camelyon.