Coaches and athletes need to understand the kinematics and dynamics of karate kicks to improve the training process and results. The research was aimed at studying the automatic recognition of punches in karate using only linear acceleration sensors. Accelerometers were part of the Inertial Measurement Units (IMUs), which were attached to the left and right wrist of the athlete. To develop a model of punches, highly qualified athletes with 3-7 years of karate experience participated in the research. We analyzed the acceleration fields of various karate punches: Yun Tsuki, Mawashi Tsuki, Age of Tsuki, Uraken. A simpler approach to extracting features without calculating their statistical characteristics is proposed. To solve the classification problem, various architectures of convolutional neural networks are used: multilayer perceptron, 1-and 2-dimension Convolution Networks. Since the recognition of punches was carried out in the conditions of a shadow fight, in addition to the recognition of punches, another output parameter was introducedmovement without punches. Studies have shown a high level of punch recognition based on the developed models. The multi-class accuracy value is 0.96, and the average F1 value is 0.97 for five different punch classes. Thus, the proposed approach is more suitable for practical implementation in automatic learning systems.

The article considers the solution of the problem of fuzzy measurement of the coordinates of small objects occupying less than 1% of the area in highresolution images. The EfficientNet and MobileNet classifiers from the Tensorflow library, pre-trained on ImageNet data, are used as the basis of the algorithm for extracting features of small objects. Next, the feature map from the last layers of the neural network is fed to the input of a two-layer perceptron, which implements classification by the feature vector in depth, for each element along the X and Y coordinate axes. The results are interpreted as a measure of the presence of an object in the receptive field in the original image. Thus, the authors have significantly simplified the architecture of the solution, while achieving acceptable indicators of accuracy and precision – 68 and 85%, respectively, the inference time on mobile platforms is less than 1 second.