4results about How to "Improve task performance" patented technology

A multi-mode autonomous underwater robot, which is a whole body streamlined structure of a rotary body, is composed of a bow load cabin section, a buoyancy adjusting cabin section, a battery cabin section, a navigation cabin section, a control cabin section, a steering cabin section, a propelling cabin section and an antenna in series by sequentially arranging, and is sealed and connected between adjacent cabin sections by a hoop. In the aspects of hardware and structure, the multi-mode autonomous underwater robot has the characteristics of modularity and easy expansion. When the function needs to be expanded, the load sensor in the bow load cabin section can be replaced as needed, all cabin sections adopt the same electrical interface scheme, and the function expansion can also be realized by additionally installing an additional sensor load cabin section. The multi-mode autonomous underwater robot realizes the function compatibility of an AUV, a floating buoy, an ARGO buoy and an underwater glider on the same underwater robot, can realize five operation modes of a water surface navigation mode, an underwater navigation mode, a floating buoy mode, an ARGO buoy mode and an underwater glider mode, the operation modes can be freely switched, and the application range and task capability of the underwater robot are greatly expanded.

This invention belongs to the field of radar information processing technology and discloses a pre-training method for a multimodal model of radar target detection, tracking, and recognition. The invention includes collecting model pre-training data; converting the collected radar information processing data from various models into a unified data format through data preprocessing to serve as input data for the pre-trained model; achieving simultaneous input of multimodal data, including images, tracks, and category text, through a unified pre-training framework, and completing joint representation learning of the input data by the pre-training framework; and designing a multi-task joint learning optimization algorithm to achieve parallel training of radar target detection, tracking, and recognition tasks. This invention achieves integrated processing of radar target detection, tracking, and recognition tasks by establishing a multi-task learning mechanism under a unified model training framework, effectively improving the overall performance of radar information processing. Simultaneously, by enhancing versatility and generalization, the model is applicable to intelligent processing tasks of various radar models.

This invention provides a lightweight adaptation method and system for large models. By extracting the semantic core from educational data, domain topic vector clusters are obtained, and pseudo-instruction-response pairs are generated to expand the training set based on these clusters. In terms of model structure, a low-rank adapter module is inserted into the transformer layer of the pre-trained large model, and the semantic direction of the topic vector clusters is extracted through singular value decomposition. The adapter parameters are initialized, and a weighted strategy is adopted during training: the prediction entropy and cross-entropy loss of each sample are calculated, and sample weights and gradient adjustment coefficients are generated based on these. The gradient is scaled and updated, and orthogonality constraints are applied, with the strength of the constraints adjusted by the sample weights.

The application relates to a Hilbert fractal curve-based image convolution method and system, comprising the following steps: 1) performing a padding operation on an input image according to a convolution output size mode, for example, the input size is equal to the output mode, and a padding operation needs to be performed on the input edge; 2) block index and rearrangement: performing image block reading and rearrangement operations on the padded input image according to the index sequence number of the Hilbert fractal; 3) cross-correlation operation: performing convolution cross-correlation operation on the rearranged image block with a step equal to the convolution kernel size; 4) local feature aggregation: performing convolution cross-correlation operation on the output of step 3 with a step of 1 by using a deep separable one-dimensional convolution; 5) restoring the two-dimensional structure of the input: performing reverse rearrangement operation on the output of step 4 according to the index of the Hilbert fractal, and outputting. Compared with the existing image convolution operation method, the application is simple, has strong universality and has certain theoretical advancement.