4results about How to "Effective focus" patented technology

The application provides an open vocabulary target detection method and system, which comprises the following steps: integrating a learnable prompt anchor module in a multi-scale feature layer of a preset cross-modal feature fusion network; injecting each prompt anchor point vector into a corresponding feature map through residual connection to dynamically correct the multi-scale semantic bias of the feature map; in the model training process, using a frozen CLIP visual encoder as a teacher model and the preset cross-modal feature fusion network as a student model to train a corresponding small target region mask, and calculating the distillation loss of the dynamically corrected feature map according to the small target region mask; outputting the multi-scale features corresponding to the feature map according to the distillation loss, and performing boundary box regression and open vocabulary classification on the multi-scale features through a preset detection head to complete the corresponding target detection. The application can effectively improve the detection efficiency of small region targets.

The utility model provides a kind of hidden antiskid acrylic bathtub belongs to the field.It solves the inner wall of traditional bathtub is often too smooth, and user is difficult to find the position that can lend force to steady posture or buffer impact in the moment of slipping, in the sliding process without any obstruction, body will be uncontrollably directly impact on bathtub wall, and further easily lead to the problem of user being injured in different degrees.This hidden antiskid acrylic bathtub includes cylinder body and faucet for water discharge installed at the edge of cylinder body, a plurality of knobs for controlling the water outlet of faucet are provided on one side of faucet, water outlet hole for water outlet is provided in cylinder body, two symmetrical anti-skid grooves are provided on the inner wall of bathtub, and the anti-skid grooves are arc-shaped.The utility model has the advantages of safety and anti-skid.

This invention discloses an autofocus method, apparatus, electronic device, and storage medium, belonging to the field of autofocus technology. The autofocus method includes: adjusting the focal length within a first focusing range according to a first preset step size, acquiring a first image captured by the camera at each focal length, and determining an effective focusing area and a second focusing range based on the first image; adjusting the focal length within a second focusing range according to a second preset step size, acquiring a second image captured by the camera at each focal length, and determining the optimal focusing position based on the sharpness evaluation value of the effective focusing area in the second image; wherein the first preset step size is greater than the second preset step size. This invention performs focusing in two stages: the first stage uses a large step size to filter the effective focusing area and narrow the focusing range, and the second stage uses a small step size for focusing and uses fewer but more effective areas for focusing, thus achieving higher operating efficiency.

The invention relates to the technical field of radars, and particularly discloses a low-slow small target micro-motion feature extraction algorithm based on STFT-LSTM. The method comprises the following steps: converting an original signal into time-frequency domain representation through STFT, and constructing a feature matrix with time sequence relevance; a bidirectional LSTM network is introduced to carry out time sequence dependence modeling, adaptive fusion of a feature space is realized in combination with a multi-head attention mechanism, and the calculation efficiency is remarkably improved through a learnable weight distribution strategy. And designing a feature visualization analysis module to verify the performance of the model, quantitatively evaluating the mapping relation between the feature parameters and the classification accuracy, and finally forming a dynamic optimization closed loop. According to the technical scheme, a new technical path is provided for multi-mode radar echo signal processing, the precision and efficiency of feature extraction are effectively improved, and the method is suitable for a target recognition task in a broadband complex electromagnetic environment.